2557 lines
105 KiB
PowerShell
2557 lines
105 KiB
PowerShell
param(
|
|
[ValidateSet("Status", "Recover", "StartVMs", "PublicSmoke", "HarborRepair", "AwoooRepair", "Perf", "LoadShed", "SelfCheck", "BackupCheck", "ProviderFreshness", "ControlTick")]
|
|
[string]$Mode = "Status",
|
|
[switch]$ControlledApply,
|
|
[string]$ConfigPath = "C:\Wooo\Agent99\config\agent99.config.json",
|
|
[string]$EvidenceDir = "C:\Wooo\Agent99\evidence"
|
|
)
|
|
|
|
$ErrorActionPreference = "Continue"
|
|
New-Item -ItemType Directory -Force -Path $EvidenceDir | Out-Null
|
|
|
|
$stamp = Get-Date -Format "yyyyMMdd-HHmmss"
|
|
$logPath = Join-Path $EvidenceDir "agent99-$Mode-$stamp.log"
|
|
$jsonPath = Join-Path $EvidenceDir "agent99-$Mode-$stamp.json"
|
|
|
|
function Write-AgentLog {
|
|
param([string]$Message)
|
|
$line = "$(Get-Date -Format o) $Message"
|
|
$line | Tee-Object -FilePath $logPath -Append | Out-Null
|
|
}
|
|
|
|
function Load-Config {
|
|
if (-not (Test-Path $ConfigPath)) {
|
|
throw "Config not found: $ConfigPath"
|
|
}
|
|
Get-Content $ConfigPath -Raw | ConvertFrom-Json
|
|
}
|
|
|
|
$Config = Load-Config
|
|
$SshUser = if ($Config.sshUser) { $Config.sshUser } else { "wooo" }
|
|
$script:Events = @()
|
|
$script:TelegramAttempts = @()
|
|
|
|
function Get-HostSshUser {
|
|
param([string]$TargetHost)
|
|
if ($Config.sshUsers -and $Config.sshUsers.PSObject.Properties[$TargetHost]) {
|
|
return [string]$Config.sshUsers.PSObject.Properties[$TargetHost].Value
|
|
}
|
|
$SshUser
|
|
}
|
|
|
|
function Send-AgentTelegramRelay {
|
|
param(
|
|
[string]$Message,
|
|
[object]$Relay,
|
|
[string]$ChatIdOverride = ""
|
|
)
|
|
|
|
if (-not ($Relay -and $Relay.enabled)) {
|
|
return [pscustomobject]@{ ok = $false; error = "relay_disabled" }
|
|
}
|
|
|
|
$relayHost = if ($Relay.host) { [string]$Relay.host } else { "192.168.0.110" }
|
|
$container = if ($Relay.container) { [string]$Relay.container } else { "stockplatform-v2-api-1" }
|
|
if ($container -notmatch "^[A-Za-z0-9_.-]+$") {
|
|
return [pscustomobject]@{ ok = $false; error = "invalid_relay_container" }
|
|
}
|
|
|
|
$messageBytes = [Text.Encoding]::UTF8.GetBytes($Message)
|
|
$messageB64 = [Convert]::ToBase64String($messageBytes)
|
|
$chatOverrideB64 = if ($ChatIdOverride) { [Convert]::ToBase64String([Text.Encoding]::UTF8.GetBytes($ChatIdOverride)) } else { "" }
|
|
$python = 'import base64, os, sys, urllib.parse, urllib.request
|
|
text = base64.b64decode(sys.argv[2]).decode()
|
|
chat_override = base64.b64decode(sys.argv[3]).decode() if len(sys.argv) > 3 and sys.argv[3] else ""
|
|
token = os.environ.get("TELEGRAM_BOT_TOKEN")
|
|
chat = chat_override or os.environ.get("TELEGRAM_CHAT_ID")
|
|
assert token and chat
|
|
data = urllib.parse.urlencode({"chat_id": chat, "text": text, "disable_web_page_preview": "true"}).encode()
|
|
urllib.request.urlopen("https://api.telegram.org/bot" + token + "/sendMessage", data=data, timeout=15).read()
|
|
print("sent_ok")
|
|
'
|
|
$pythonB64 = [Convert]::ToBase64String([Text.Encoding]::UTF8.GetBytes($python))
|
|
$bootstrap = 'import base64,sys;exec(base64.b64decode(sys.argv[1]))'
|
|
$remoteCommand = "docker exec $container python3 -c '$bootstrap' $pythonB64 $messageB64 $chatOverrideB64"
|
|
$run = Invoke-SshText $relayHost $remoteCommand 45 1
|
|
|
|
[pscustomobject]@{
|
|
ok = [bool]($run.ok -and $run.output -match "sent_ok")
|
|
host = $relayHost
|
|
container = $container
|
|
exitCode = $run.exitCode
|
|
chatOverride = if ($ChatIdOverride) { "telegram-origin" } else { $null }
|
|
output = $run.output
|
|
error = if ($run.ok) { $null } else { $run.output }
|
|
}
|
|
}
|
|
|
|
function Format-AgentTelegramText {
|
|
param(
|
|
[string]$Severity,
|
|
[string]$EventType,
|
|
[string]$Message,
|
|
[object]$Data = $null
|
|
)
|
|
|
|
$severityLabel = switch ($Severity) {
|
|
"critical" { "CRITICAL" }
|
|
"warning" { "WARNING" }
|
|
"info" { "INFO" }
|
|
default { $Severity }
|
|
}
|
|
$lines = @()
|
|
$lines += "Agent99 Alert: $severityLabel"
|
|
|
|
if ($EventType -match "^performance_") {
|
|
$hostName = if ($Data -and $Data.PSObject.Properties["host"]) { $Data.host } else { "unknown" }
|
|
$reasons = if ($Data -and $Data.PSObject.Properties["reasons"]) { @($Data.reasons) } else { @() }
|
|
$disk = if ($Data -and $Data.PSObject.Properties["diskUsedPercent"]) { $Data.diskUsedPercent } else { $null }
|
|
$load = if ($Data -and $Data.PSObject.Properties["loadPerCore"]) { $Data.loadPerCore } else { $null }
|
|
$mem = if ($Data -and $Data.PSObject.Properties["memAvailablePercent"]) { $Data.memAvailablePercent } else { $null }
|
|
$lines += "Event: host performance"
|
|
$lines += "Host: $hostName"
|
|
if ($reasons -contains "disk_used_warning") {
|
|
$lines += "Reason: disk usage is $disk percent. Warning threshold reached; critical threshold is 95 percent."
|
|
$lines += "Action: Agent99 will run allowlisted disk remediation and suppress duplicates for 30 minutes."
|
|
} elseif ($reasons -contains "disk_used_high") {
|
|
$lines += "Reason: disk usage is $disk percent. Critical threshold reached."
|
|
$lines += "Action: Agent99 will run allowlisted cleanup first; reboot is not the first response."
|
|
} elseif ($reasons -contains "load_per_core_warning" -or $reasons -contains "load_per_core_critical") {
|
|
$lines += "Reason: CPU load/core=$load."
|
|
$lines += "Action: Agent99 will use allowlisted load reduction only; no arbitrary service stop."
|
|
} elseif ($reasons -contains "memory_available_low") {
|
|
$lines += "Reason: available memory is $mem percent."
|
|
$lines += "Action: Agent99 will inspect memory pressure before allowlisted remediation."
|
|
} else {
|
|
$lines += "Reason: $($reasons -join ', ')"
|
|
$lines += "Action: check Agent99 evidence for full readback."
|
|
}
|
|
$lines += "Current: load/core=$load, memAvail=$mem percent, disk=$disk percent."
|
|
} elseif ($EventType -match "^backup_") {
|
|
$targetName = if ($Data -and $Data.PSObject.Properties["name"]) { $Data.name } else { "unknown" }
|
|
$reason = if ($Data -and $Data.PSObject.Properties["reason"]) { $Data.reason } else { $EventType }
|
|
$age = if ($Data -and $Data.PSObject.Properties["ageMinutes"]) { $Data.ageMinutes } else { $null }
|
|
$maxAge = if ($Data -and $Data.PSObject.Properties["maxAgeMinutes"]) { $Data.maxAgeMinutes } else { $null }
|
|
$lines += "Event: backup health"
|
|
$lines += "Target: $targetName"
|
|
$lines += "Reason: $reason"
|
|
if ($age -ne $null -or $maxAge -ne $null) {
|
|
$lines += "Freshness: age=$age minutes / max=$maxAge minutes"
|
|
}
|
|
$lines += "Action: inspect /backup/status, /backup/logs, and snapshot metadata only; do not read backup contents or secrets."
|
|
} elseif ($EventType -eq "provider_freshness_triage") {
|
|
$candidateId = if ($Data -and $Data.PSObject.Properties["candidateId"]) { $Data.candidateId } else { "unknown" }
|
|
$targetName = if ($Data -and $Data.PSObject.Properties["target"]) { $Data.target } else { "provider_freshness_signal" }
|
|
$hitCount = if ($Data -and $Data.PSObject.Properties["hitCount"]) { $Data.hitCount } else { 0 }
|
|
$missing = if ($Data -and $Data.PSObject.Properties["missingFields"]) { @($Data.missingFields) } else { @() }
|
|
$candidatePath = if ($Data -and $Data.PSObject.Properties["candidatePath"]) { $Data.candidatePath } else { $null }
|
|
$readback = if ($Data -and $Data.PSObject.Properties["verifierReadback"]) { $Data.verifierReadback } else { $null }
|
|
$status = if ($Data -and $Data.PSObject.Properties["status"]) { $Data.status } else { "unknown" }
|
|
$lines += "Event: provider freshness auto triage"
|
|
$lines += "Target: $targetName"
|
|
$lines += "Status: $status"
|
|
$lines += "Evidence hits: $hitCount"
|
|
if ($readback) {
|
|
$lines += "Direct readback: ok=$($readback.ok), sourceEvents=$($readback.sourceEventTotal), latest=$($readback.latestReceivedAt)"
|
|
}
|
|
if (@($missing).Count -gt 0) {
|
|
$lines += "Still missing: $($missing -join ', ')"
|
|
} else {
|
|
$lines += "Missing: direct readback fields completed"
|
|
}
|
|
$lines += "Action: candidate and KM/RAG were written; duplicate same-state alerts are suppressed."
|
|
$lines += "Guard: no provider switch, paid model call, env change, or reload."
|
|
if ($candidatePath) {
|
|
$lines += "Evidence: $candidatePath"
|
|
}
|
|
} elseif ($EventType -match "^load_shed_") {
|
|
$actionName = if ($Data -and $Data.PSObject.Properties["name"]) { $Data.name } else { "unknown" }
|
|
$hostName = if ($Data -and $Data.PSObject.Properties["host"]) { $Data.host } else { "unknown" }
|
|
$ok = if ($Data -and $Data.PSObject.Properties["ok"]) { $Data.ok } else { $null }
|
|
$executionOk = if ($Data -and $Data.PSObject.Properties["executionOk"]) { $Data.executionOk } else { $null }
|
|
$postVerifyOk = if ($Data -and $Data.PSObject.Properties["postVerifyOk"]) { $Data.postVerifyOk } else { $null }
|
|
$exitCode = if ($Data -and $Data.PSObject.Properties["exitCode"]) { $Data.exitCode } else { $null }
|
|
$rollback = if ($Data -and $Data.PSObject.Properties["rollback"]) { $Data.rollback } else { $null }
|
|
$lines += "Event: controlled remediation"
|
|
$lines += "Host: $hostName"
|
|
$lines += "Action: $actionName"
|
|
if ($ok -ne $null) {
|
|
$lines += "Result: ok=$ok exitCode=$exitCode"
|
|
if ($executionOk -ne $null -or $postVerifyOk -ne $null) {
|
|
$lines += "Verifier: executionOk=$executionOk postVerifyOk=$postVerifyOk"
|
|
}
|
|
} else {
|
|
$lines += "Result: check Agent99 evidence."
|
|
}
|
|
if ($rollback) {
|
|
$lines += "Rollback: $rollback"
|
|
}
|
|
} elseif ($EventType -match "^operator_command_") {
|
|
$commandId = if ($Data -and $Data.PSObject.Properties["id"]) { $Data.id } else { "unknown" }
|
|
$modeName = if ($Data -and $Data.PSObject.Properties["mode"]) { $Data.mode } else { "unknown" }
|
|
$source = if ($Data -and $Data.PSObject.Properties["source"]) { $Data.source } else { "unknown" }
|
|
$instruction = if ($Data -and $Data.PSObject.Properties["instruction"]) { $Data.instruction } else { $null }
|
|
$exitCode = if ($Data -and $Data.PSObject.Properties["exitCode"]) { $Data.exitCode } else { $null }
|
|
$evidencePath = if ($Data -and $Data.PSObject.Properties["evidence"]) { $Data.evidence } else { $null }
|
|
$lines += "Event: Agent99 automation result"
|
|
$lines += "Mode: $modeName"
|
|
$lines += "Source: $source"
|
|
$lines += "Result: exitCode=$exitCode"
|
|
if ($Data -and $Data.PSObject.Properties["problem"]) {
|
|
$problem = $Data.problem
|
|
if ($problem -and $problem.PSObject.Properties["occurrences"]) {
|
|
$lines += "Problem count: occurrences=$($problem.occurrences), resolved=$($problem.resolved), failed=$($problem.failed)"
|
|
}
|
|
}
|
|
if ($instruction -and $exitCode -ne 0) {
|
|
$lines += "Failure instruction: $instruction"
|
|
}
|
|
if ($evidencePath) {
|
|
$lines += "Evidence: $evidencePath"
|
|
}
|
|
} elseif ($EventType -eq "agent_control_tick") {
|
|
$dashboardPath = if ($Data -and $Data.PSObject.Properties["dashboardPath"]) { $Data.dashboardPath } else { $null }
|
|
$processed = if ($Data -and $Data.PSObject.Properties["processedCount"]) { $Data.processedCount } else { 0 }
|
|
$pending = if ($Data -and $Data.PSObject.Properties["pendingCount"]) { $Data.pendingCount } else { 0 }
|
|
$lines += "Event: Agent99 control loop"
|
|
$lines += "Dashboard: $dashboardPath"
|
|
$lines += "Queue: processed=$processed pending=$pending"
|
|
} elseif ($EventType -match "^ai_service_") {
|
|
$serviceName = if ($Data -and $Data.PSObject.Properties["name"]) { $Data.name } else { "unknown" }
|
|
$lines += "Event: AI service health"
|
|
$lines += "Service: $serviceName"
|
|
$lines += "Action: inspect health endpoint and service/container state before controlled repair."
|
|
} elseif ($EventType -eq "agent_selfcheck") {
|
|
$lines += "Event: Agent99 self health"
|
|
$lines += "Summary: $Message"
|
|
$lines += "Action: inspect task state, evidence freshness, and Telegram delivery; reboot is not the first response."
|
|
} else {
|
|
$lines += "Event: $EventType"
|
|
$lines += "Summary: $Message"
|
|
}
|
|
|
|
$lines += "Time: $(Get-Date -Format 'yyyy-MM-dd HH:mm:ss zzz')"
|
|
$text = ($lines -join "`n")
|
|
if ($text.Length -gt 3500) {
|
|
$text = $text.Substring(0, 3500) + "`n...(truncated)"
|
|
}
|
|
$text
|
|
}
|
|
|
|
function Send-AgentTelegram {
|
|
param(
|
|
[string]$Severity,
|
|
[string]$EventType,
|
|
[string]$Message,
|
|
[object]$Data = $null
|
|
)
|
|
|
|
$telegram = $Config.telegram
|
|
$enabled = ($telegram -and $telegram.enabled)
|
|
$botTokenEnv = if ($telegram.botTokenEnv) { $telegram.botTokenEnv } else { "AGENT99_TELEGRAM_BOT_TOKEN" }
|
|
$chatIdEnv = if ($telegram.chatIdEnv) { $telegram.chatIdEnv } else { "AGENT99_TELEGRAM_CHAT_ID" }
|
|
$tokenEnvAliases = @($botTokenEnv)
|
|
if ($telegram.botTokenEnvAliases) { $tokenEnvAliases += @($telegram.botTokenEnvAliases) }
|
|
$chatIdEnvAliases = @($chatIdEnv)
|
|
if ($telegram.chatIdEnvAliases) { $chatIdEnvAliases += @($telegram.chatIdEnvAliases) }
|
|
|
|
function Get-AgentEnvSecret {
|
|
param([string[]]$Names)
|
|
foreach ($name in ($Names | Where-Object { $_ } | Select-Object -Unique)) {
|
|
foreach ($scope in @("Machine", "User", "Process")) {
|
|
$value = [Environment]::GetEnvironmentVariable($name, $scope)
|
|
if ($value) {
|
|
return [pscustomobject]@{ value = $value; name = $name; scope = $scope }
|
|
}
|
|
}
|
|
}
|
|
return $null
|
|
}
|
|
|
|
$tokenSecret = Get-AgentEnvSecret $tokenEnvAliases
|
|
$chatSecret = Get-AgentEnvSecret $chatIdEnvAliases
|
|
$token = if ($tokenSecret) { $tokenSecret.value } else { $null }
|
|
$chatId = if ($chatSecret) { $chatSecret.value } else { $null }
|
|
$chatIdOverride = if ($Data -and $Data.PSObject.Properties["replyChatId"]) { [string]$Data.replyChatId } else { "" }
|
|
|
|
$attempt = [pscustomobject]@{
|
|
timestamp = (Get-Date -Format o)
|
|
enabled = [bool]$enabled
|
|
eventType = $EventType
|
|
severity = $Severity
|
|
configured = [bool]($token -and $chatId)
|
|
tokenEnv = if ($tokenSecret) { $tokenSecret.name } else { $null }
|
|
chatIdEnv = if ($chatSecret) { $chatSecret.name } else { $null }
|
|
chatOverride = if ($chatIdOverride) { "telegram-origin" } else { $null }
|
|
relay = $null
|
|
sent = $false
|
|
error = $null
|
|
}
|
|
|
|
if (-not $enabled) {
|
|
$attempt.error = "telegram_disabled"
|
|
$script:TelegramAttempts += $attempt
|
|
return
|
|
}
|
|
$text = Format-AgentTelegramText $Severity $EventType $Message $Data
|
|
if (-not ($token -and ($chatId -or $chatIdOverride))) {
|
|
$relayResult = Send-AgentTelegramRelay $text $telegram.relay $chatIdOverride
|
|
$attempt.relay = $relayResult
|
|
if ($relayResult.ok) {
|
|
$attempt.sent = $true
|
|
$attempt.error = $null
|
|
} else {
|
|
$attempt.error = "telegram_not_configured"
|
|
}
|
|
$script:TelegramAttempts += $attempt
|
|
return
|
|
}
|
|
|
|
try {
|
|
Invoke-RestMethod -Method Post -Uri "https://api.telegram.org/bot$token/sendMessage" -Body @{
|
|
chat_id = if ($chatIdOverride) { $chatIdOverride } else { $chatId }
|
|
text = $text
|
|
disable_web_page_preview = "true"
|
|
} | Out-Null
|
|
$attempt.sent = $true
|
|
} catch {
|
|
$attempt.error = $_.Exception.Message
|
|
}
|
|
|
|
$script:TelegramAttempts += $attempt
|
|
}
|
|
|
|
function Record-AgentEvent {
|
|
param(
|
|
[string]$EventType,
|
|
[string]$Severity,
|
|
[string]$Message,
|
|
[object]$Data = $null,
|
|
[switch]$Alert,
|
|
[switch]$NoAlert
|
|
)
|
|
|
|
Write-AgentLog "event type=$EventType severity=$Severity message=$Message"
|
|
$eventData = $Data
|
|
if ($Data -and $Data.PSObject.Properties["replyChatId"]) {
|
|
$eventData = $Data | Select-Object * -ExcludeProperty replyChatId
|
|
$eventData | Add-Member -MemberType NoteProperty -Name replyChat -Value "telegram-origin" -Force
|
|
}
|
|
$event = [pscustomobject]@{
|
|
timestamp = (Get-Date -Format o)
|
|
type = $EventType
|
|
severity = $Severity
|
|
message = $Message
|
|
data = $eventData
|
|
}
|
|
$script:Events += $event
|
|
|
|
if ($NoAlert) {
|
|
return
|
|
}
|
|
|
|
$alertAll = ($Config.telegram -and $Config.telegram.alertAllOperations)
|
|
$alertInfo = ($Config.telegram -and $Config.telegram.alertInfoEvents)
|
|
$alertOperatorCommands = $true
|
|
if ($Config.telegram -and $Config.telegram.PSObject.Properties["alertOperatorCommands"]) {
|
|
$alertOperatorCommands = [bool]$Config.telegram.alertOperatorCommands
|
|
}
|
|
$operatorCommandAlert = [bool]($alertOperatorCommands -and $EventType -match "^operator_command_")
|
|
if ($operatorCommandAlert -and $Data) {
|
|
$operatorOk = $false
|
|
$operatorMode = ""
|
|
$operatorId = ""
|
|
$operatorOkValue = $null
|
|
$operatorExitCode = $null
|
|
if ($Data -is [System.Collections.IDictionary]) {
|
|
if ($Data.Contains("mode")) { $operatorMode = [string]$Data["mode"] }
|
|
if ($Data.Contains("id")) { $operatorId = [string]$Data["id"] }
|
|
if ($Data.Contains("ok")) { $operatorOkValue = $Data["ok"] }
|
|
if ($Data.Contains("exitCode")) { $operatorExitCode = $Data["exitCode"] }
|
|
} else {
|
|
$modeProp = $Data.PSObject.Properties.Match("mode") | Select-Object -First 1
|
|
$idProp = $Data.PSObject.Properties.Match("id") | Select-Object -First 1
|
|
$okProp = $Data.PSObject.Properties.Match("ok") | Select-Object -First 1
|
|
$exitCodeProp = $Data.PSObject.Properties.Match("exitCode") | Select-Object -First 1
|
|
if ($modeProp) { $operatorMode = [string]$modeProp.Value }
|
|
if ($idProp) { $operatorId = [string]$idProp.Value }
|
|
if ($okProp) { $operatorOkValue = $okProp.Value }
|
|
if ($exitCodeProp) { $operatorExitCode = $exitCodeProp.Value }
|
|
}
|
|
if ($null -ne $operatorOkValue -and [string]$operatorOkValue -match "^(?i:true|1)$") { $operatorOk = $true }
|
|
$parsedExitCode = -999
|
|
if ($null -ne $operatorExitCode -and [int]::TryParse([string]$operatorExitCode, [ref]$parsedExitCode) -and $parsedExitCode -eq 0) { $operatorOk = $true }
|
|
$alertSuccessfulOperatorCommands = $false
|
|
if ($Config.telegram -and $Config.telegram.PSObject.Properties["alertSuccessfulOperatorCommands"]) {
|
|
$alertSuccessfulOperatorCommands = [bool]$Config.telegram.alertSuccessfulOperatorCommands
|
|
}
|
|
if ($operatorOk -and -not $alertSuccessfulOperatorCommands) {
|
|
Write-AgentLog "telegram_operator_success_suppressed type=$EventType mode=$operatorMode id=$operatorId"
|
|
$operatorCommandAlert = $false
|
|
}
|
|
}
|
|
$shouldAlert = ($Severity -in @("warning", "critical")) -or $operatorCommandAlert -or ($Alert -and $Severity -ne "info") -or ($alertAll -and ($Severity -ne "info" -or $alertInfo))
|
|
if ($shouldAlert) {
|
|
$dedupeMinutes = 30
|
|
if ($Config.telegram -and $Config.telegram.PSObject.Properties["dedupeMinutes"]) {
|
|
$dedupeMinutes = [int]$Config.telegram.dedupeMinutes
|
|
}
|
|
$dedupeParts = @($EventType, $Severity)
|
|
if ($Data) {
|
|
foreach ($propName in @("host", "name", "path", "reason", "target", "status", "mode", "alertKind", "source")) {
|
|
if ($Data.PSObject.Properties[$propName]) {
|
|
$dedupeParts += "$propName=$($Data.PSObject.Properties[$propName].Value)"
|
|
}
|
|
}
|
|
if ($Data.PSObject.Properties["reasons"]) {
|
|
$dedupeParts += "reasons=$(@($Data.reasons) -join ',')"
|
|
}
|
|
if ($Data.PSObject.Properties["missingFields"]) {
|
|
$dedupeParts += "missing=$(@($Data.missingFields) -join ',')"
|
|
}
|
|
if ($Data.PSObject.Properties["problem"] -and $Data.problem -and $Data.problem.PSObject.Properties["key"]) {
|
|
$dedupeParts += "problem=$($Data.problem.key)"
|
|
}
|
|
}
|
|
if ($dedupeParts.Count -le 2) {
|
|
$normalizedMessage = $Message
|
|
foreach ($pattern in @("candidate=[^ ]+", "evidence=[^ ]+", "loadPerCore=[^ ]+", "memAvailPct=[^ ]+", "diskUsedPct=[^ ]+", "ageMinutes=[^ ]+")) {
|
|
$normalizedMessage = $normalizedMessage -replace $pattern, ($pattern.Split("=")[0] + "=<value>")
|
|
}
|
|
$dedupeParts += $normalizedMessage
|
|
}
|
|
$dedupeSource = ($dedupeParts -join "|")
|
|
$dedupeHashBytes = [Security.Cryptography.SHA256]::Create().ComputeHash([Text.Encoding]::UTF8.GetBytes($dedupeSource))
|
|
$dedupeHash = [BitConverter]::ToString($dedupeHashBytes).Replace("-", "").ToLowerInvariant()
|
|
$dedupePath = Join-Path $EvidenceDir "telegram-dedupe-$dedupeHash.marker"
|
|
if ($dedupeMinutes -gt 0 -and (Test-Path $dedupePath)) {
|
|
$ageMinutes = ((Get-Date) - (Get-Item $dedupePath).LastWriteTime).TotalMinutes
|
|
if ($ageMinutes -lt $dedupeMinutes) {
|
|
Write-AgentLog "telegram_dedupe_skip type=$EventType severity=$Severity ageMinutes=$([math]::Round($ageMinutes, 2)) cooldown=$dedupeMinutes"
|
|
return
|
|
}
|
|
}
|
|
Set-Content -Path $dedupePath -Value (Get-Date -Format o) -Encoding UTF8
|
|
Send-AgentTelegram $Severity $EventType $Message $Data
|
|
}
|
|
}
|
|
|
|
function Invoke-SshText {
|
|
param(
|
|
[string]$TargetHost,
|
|
[string]$Command,
|
|
[int]$TimeoutSeconds = 12,
|
|
[int]$Retries = 2,
|
|
[string]$User = $null
|
|
)
|
|
|
|
$last = $null
|
|
$userToUse = if ($User) { $User } else { Get-HostSshUser $TargetHost }
|
|
for ($attempt = 1; $attempt -le $Retries; $attempt++) {
|
|
$args = @(
|
|
"-o", "BatchMode=yes",
|
|
"-o", "StrictHostKeyChecking=accept-new",
|
|
"-o", "NumberOfPasswordPrompts=0",
|
|
"-o", "ConnectTimeout=$TimeoutSeconds",
|
|
"-o", "ServerAliveInterval=20",
|
|
"-o", "ServerAliveCountMax=3",
|
|
"-l", $userToUse,
|
|
$TargetHost,
|
|
$Command
|
|
)
|
|
|
|
try {
|
|
$id = [guid]::NewGuid().ToString("N")
|
|
$stdoutPath = Join-Path $env:TEMP "agent99-ssh-$id.out"
|
|
$stderrPath = Join-Path $env:TEMP "agent99-ssh-$id.err"
|
|
$process = Start-Process -FilePath "ssh.exe" -ArgumentList $args -NoNewWindow -RedirectStandardOutput $stdoutPath -RedirectStandardError $stderrPath -PassThru
|
|
|
|
if (-not $process.WaitForExit($TimeoutSeconds * 1000)) {
|
|
$process.Kill()
|
|
$process.WaitForExit(2000) | Out-Null
|
|
Remove-Item $stdoutPath, $stderrPath -Force -ErrorAction SilentlyContinue | Out-Null
|
|
$last = [pscustomobject]@{
|
|
ok = $false
|
|
exitCode = -2
|
|
output = "timeout_after_${TimeoutSeconds}s"
|
|
}
|
|
} else {
|
|
$process.WaitForExit() | Out-Null
|
|
$process.Refresh()
|
|
$stdout = if (Test-Path $stdoutPath) { Get-Content $stdoutPath -Raw } else { "" }
|
|
$stderr = if (Test-Path $stderrPath) { Get-Content $stderrPath -Raw } else { "" }
|
|
Remove-Item $stdoutPath, $stderrPath -Force -ErrorAction SilentlyContinue | Out-Null
|
|
$text = (@($stdout, $stderr) | Where-Object { $_ }) -join "`n"
|
|
$exitCode = $process.ExitCode
|
|
if ($null -eq $exitCode) {
|
|
$exitCode = if ($stderr) { -3 } else { 0 }
|
|
}
|
|
$last = [pscustomobject]@{
|
|
ok = ($exitCode -eq 0)
|
|
exitCode = $exitCode
|
|
output = $text
|
|
}
|
|
}
|
|
} catch {
|
|
$last = [pscustomobject]@{
|
|
ok = $false
|
|
exitCode = -1
|
|
output = $_.Exception.Message
|
|
}
|
|
}
|
|
|
|
if ($last.ok) {
|
|
if ($attempt -gt 1) {
|
|
Write-AgentLog "ssh_retry_recovered host=$TargetHost attempt=$attempt"
|
|
}
|
|
return $last
|
|
}
|
|
|
|
if ($attempt -lt $Retries) {
|
|
Write-AgentLog "ssh_retry host=$TargetHost attempt=$attempt exit=$($last.exitCode)"
|
|
Start-Sleep -Seconds 2
|
|
}
|
|
}
|
|
|
|
$last
|
|
}
|
|
|
|
function Quote-ShSingle {
|
|
param([string]$Text)
|
|
"'" + $Text.Replace("'", "'`"`"'") + "'"
|
|
}
|
|
|
|
function Invoke-HostSshText {
|
|
param(
|
|
[string]$TargetHost,
|
|
[string]$Command,
|
|
[int]$TimeoutSeconds = 30,
|
|
[int]$Retries = 1
|
|
)
|
|
|
|
$direct = Invoke-SshText $TargetHost $Command $TimeoutSeconds $Retries
|
|
if ($direct.ok -or -not $Config.k3s.jumpHost -or $TargetHost -eq $Config.k3s.jumpHost) {
|
|
$direct | Add-Member -NotePropertyName route -NotePropertyValue "direct" -Force
|
|
return $direct
|
|
}
|
|
|
|
$quotedCommand = Quote-ShSingle $Command
|
|
$targetUser = Get-HostSshUser $TargetHost
|
|
$jumpCommand = "ssh -o BatchMode=yes -o StrictHostKeyChecking=accept-new -o NumberOfPasswordPrompts=0 -o ConnectTimeout=10 -l $targetUser $TargetHost $quotedCommand"
|
|
$viaJump = Invoke-SshText $Config.k3s.jumpHost $jumpCommand $TimeoutSeconds $Retries
|
|
$viaJump | Add-Member -NotePropertyName route -NotePropertyValue "via_$($Config.k3s.jumpHost)" -Force
|
|
return $viaJump
|
|
}
|
|
|
|
function Convert-AgentDouble {
|
|
param([string]$Value)
|
|
if (-not $Value) { return $null }
|
|
try {
|
|
return [double]::Parse($Value.Replace(",", "."), [Globalization.CultureInfo]::InvariantCulture)
|
|
} catch {
|
|
return $null
|
|
}
|
|
}
|
|
|
|
function Get-PerfThresholds {
|
|
$perf = $Config.performance
|
|
[pscustomobject]@{
|
|
loadPerCoreWarning = if ($perf.loadPerCoreWarning) { [double]$perf.loadPerCoreWarning } else { 0.9 }
|
|
loadPerCoreCritical = if ($perf.loadPerCoreCritical) { [double]$perf.loadPerCoreCritical } else { 1.5 }
|
|
memoryAvailablePercentCritical = if ($perf.memoryAvailablePercentCritical) { [double]$perf.memoryAvailablePercentCritical } else { 10 }
|
|
diskUsedPercentWarning = if ($perf.diskUsedPercentWarning) { [double]$perf.diskUsedPercentWarning } else { 90 }
|
|
diskUsedPercentCritical = if ($perf.diskUsedPercentCritical) { [double]$perf.diskUsedPercentCritical } else { 95 }
|
|
}
|
|
}
|
|
|
|
function Get-HostPerformance {
|
|
param([switch]$SuppressAlerts)
|
|
|
|
$thresholds = Get-PerfThresholds
|
|
$results = @()
|
|
$command = @'
|
|
echo __AGENT99_PERF__
|
|
echo HOST=$(hostname)
|
|
echo CORES=$(nproc 2>/dev/null || getconf _NPROCESSORS_ONLN 2>/dev/null || echo 1)
|
|
uptime
|
|
echo __MEM__
|
|
free -m 2>/dev/null
|
|
echo __DISK__
|
|
df -P / 2>/dev/null
|
|
'@
|
|
|
|
foreach ($hostIp in $Config.hosts) {
|
|
$readback = Invoke-HostSshText $hostIp $command 25 1
|
|
$text = $readback.output
|
|
$cores = 1
|
|
$coresMatch = [regex]::Match($text, "CORES=(\d+)")
|
|
if ($coresMatch.Success) { $cores = [int]$coresMatch.Groups[1].Value }
|
|
|
|
$load1 = $null
|
|
$loadMatch = [regex]::Match($text, "load averages?:\s*([0-9]+(?:[\.,][0-9]+)?)")
|
|
if (-not $loadMatch.Success) {
|
|
$loadMatch = [regex]::Match($text, "load average:\s*([0-9]+(?:[\.,][0-9]+)?)")
|
|
}
|
|
if ($loadMatch.Success) {
|
|
$load1 = Convert-AgentDouble $loadMatch.Groups[1].Value
|
|
}
|
|
|
|
$memAvailablePercent = $null
|
|
$memLine = (($text -split "`n") | Where-Object { $_ -match "^\s*Mem:" } | Select-Object -First 1)
|
|
if ($memLine) {
|
|
$parts = @($memLine -split "\s+" | Where-Object { $_ })
|
|
if ($parts.Count -ge 7) {
|
|
$memTotal = Convert-AgentDouble $parts[1]
|
|
$memAvailable = Convert-AgentDouble $parts[6]
|
|
if ($memTotal -and $memTotal -gt 0 -and $null -ne $memAvailable) {
|
|
$memAvailablePercent = [math]::Round(($memAvailable / $memTotal) * 100, 2)
|
|
}
|
|
}
|
|
}
|
|
|
|
$diskUsedPercent = $null
|
|
$diskLine = (($text -split "`n") | Where-Object { $_ -match "\s+\d+%\s+/$" } | Select-Object -First 1)
|
|
if ($diskLine) {
|
|
$diskMatch = [regex]::Match($diskLine, "\s(\d+)%\s+/$")
|
|
if ($diskMatch.Success) { $diskUsedPercent = [double]$diskMatch.Groups[1].Value }
|
|
}
|
|
|
|
$topCpu = @()
|
|
|
|
$loadPerCore = if ($null -ne $load1 -and $cores -gt 0) { [math]::Round($load1 / $cores, 2) } else { $null }
|
|
$severity = "ok"
|
|
$reasons = @()
|
|
if (-not $readback.ok) {
|
|
$severity = "warning"
|
|
$reasons += "perf_readback_failed"
|
|
}
|
|
if ($null -ne $loadPerCore -and $loadPerCore -ge $thresholds.loadPerCoreCritical) {
|
|
$severity = "critical"
|
|
$reasons += "load_per_core_critical"
|
|
} elseif ($null -ne $loadPerCore -and $loadPerCore -ge $thresholds.loadPerCoreWarning -and $severity -ne "critical") {
|
|
$severity = "warning"
|
|
$reasons += "load_per_core_warning"
|
|
}
|
|
if ($null -ne $memAvailablePercent -and $memAvailablePercent -le $thresholds.memoryAvailablePercentCritical) {
|
|
$severity = "critical"
|
|
$reasons += "memory_available_low"
|
|
}
|
|
if ($null -ne $diskUsedPercent -and $diskUsedPercent -ge $thresholds.diskUsedPercentCritical) {
|
|
$severity = "critical"
|
|
$reasons += "disk_used_high"
|
|
} elseif ($null -ne $diskUsedPercent -and $diskUsedPercent -ge $thresholds.diskUsedPercentWarning -and $severity -ne "critical") {
|
|
$severity = "warning"
|
|
$reasons += "disk_used_warning"
|
|
}
|
|
|
|
$captureTopCpuOnWarning = $true
|
|
if ($Config.performance -and $Config.performance.PSObject.Properties["captureTopCpuOnWarning"]) {
|
|
$captureTopCpuOnWarning = [bool]$Config.performance.captureTopCpuOnWarning
|
|
}
|
|
$topCpuTimeoutSeconds = 10
|
|
if ($Config.performance -and $Config.performance.PSObject.Properties["topCpuTimeoutSeconds"]) {
|
|
$topCpuTimeoutSeconds = [int]$Config.performance.topCpuTimeoutSeconds
|
|
}
|
|
$shouldCaptureTopCpu = ($severity -eq "critical") -or ($severity -eq "warning" -and $captureTopCpuOnWarning)
|
|
if ($shouldCaptureTopCpu -and $readback.ok) {
|
|
$topReadback = Invoke-HostSshText $hostIp "ps -eo pid,ppid,comm,pcpu,pmem --sort=-pcpu" $topCpuTimeoutSeconds 1
|
|
if ($topReadback.ok) {
|
|
$topCpu = @($topReadback.output -split "`n" | Select-Object -First 8)
|
|
} else {
|
|
$topCpu = @("top_cpu_readback_failed: $($topReadback.output)")
|
|
}
|
|
}
|
|
|
|
$result = [pscustomobject]@{
|
|
host = $hostIp
|
|
ok = [bool]$readback.ok
|
|
route = $readback.route
|
|
severity = $severity
|
|
reasons = $reasons
|
|
cores = $cores
|
|
load1 = $load1
|
|
loadPerCore = $loadPerCore
|
|
memAvailablePercent = $memAvailablePercent
|
|
diskUsedPercent = $diskUsedPercent
|
|
topCpu = $topCpu
|
|
output = $text
|
|
}
|
|
Write-AgentLog "perf host=$hostIp severity=$severity loadPerCore=$loadPerCore memAvailPct=$memAvailablePercent diskUsedPct=$diskUsedPercent route=$($readback.route)"
|
|
if ($severity -in @("warning", "critical") -and -not $SuppressAlerts) {
|
|
Record-AgentEvent "performance_$severity" $severity "host=$hostIp loadPerCore=$loadPerCore memAvailPct=$memAvailablePercent diskUsedPct=$diskUsedPercent reasons=$($reasons -join ',')" $result -Alert
|
|
}
|
|
$results += $result
|
|
}
|
|
|
|
$results
|
|
}
|
|
|
|
function Record-PerformanceIssuesWithoutRemediation {
|
|
param(
|
|
[object[]]$Performance,
|
|
[object[]]$LoadShedding
|
|
)
|
|
|
|
foreach ($perf in @($Performance | Where-Object { $_ -and $_.severity -in @("warning", "critical") })) {
|
|
$attempts = @($LoadShedding | Where-Object { $_ -and $_.host -eq $perf.host })
|
|
$executed = @($attempts | Where-Object { $_.PSObject.Properties["skipped"] -and $_.skipped -eq $false })
|
|
if ($executed.Count -gt 0) {
|
|
continue
|
|
}
|
|
|
|
$message = "host=$($perf.host) loadPerCore=$($perf.loadPerCore) memAvailPct=$($perf.memAvailablePercent) diskUsedPct=$($perf.diskUsedPercent) reasons=$(@($perf.reasons) -join ',')"
|
|
Record-AgentEvent "performance_$($perf.severity)" $perf.severity $message $perf -Alert
|
|
}
|
|
}
|
|
|
|
function Invoke-LoadShedding {
|
|
param([object[]]$Performance)
|
|
|
|
$results = @()
|
|
$loadShedding = $Config.performance.loadShedding
|
|
if (-not ($loadShedding -and $loadShedding.enabled)) {
|
|
Write-AgentLog "load_shedding skipped enabled=false"
|
|
return $results
|
|
}
|
|
if (-not $ControlledApply) {
|
|
Write-AgentLog "load_shedding skipped controlled_apply=false"
|
|
return $results
|
|
}
|
|
|
|
foreach ($action in @($loadShedding.actions)) {
|
|
if ($action.enabled -eq $false) { continue }
|
|
$hostPerf = $Performance | Where-Object { $_.host -eq $action.host } | Select-Object -First 1
|
|
if (-not $hostPerf) { continue }
|
|
|
|
$severityRank = @{ ok = 0; warning = 1; critical = 2 }
|
|
$minSeverity = if ($action.minSeverity) { [string]$action.minSeverity } else { "critical" }
|
|
if (-not $severityRank.ContainsKey($minSeverity)) { $minSeverity = "critical" }
|
|
$hostSeverity = if ($hostPerf.severity -and $severityRank.ContainsKey([string]$hostPerf.severity)) { [string]$hostPerf.severity } else { "ok" }
|
|
if ($severityRank[$hostSeverity] -lt $severityRank[$minSeverity]) {
|
|
$results += [pscustomobject]@{ name = $action.name; host = $action.host; skipped = $true; reason = "host_below_min_severity"; hostSeverity = $hostSeverity; minSeverity = $minSeverity }
|
|
continue
|
|
}
|
|
|
|
$reasonMatches = @($action.reasonMatches)
|
|
if ($reasonMatches.Count -eq 0) { $reasonMatches = @("load_per_core_critical") }
|
|
$matchedReasons = @($hostPerf.reasons | Where-Object { $_ -in $reasonMatches })
|
|
if ($matchedReasons.Count -eq 0) {
|
|
$results += [pscustomobject]@{
|
|
name = $action.name
|
|
host = $action.host
|
|
skipped = $true
|
|
reason = "critical_reason_not_allowlisted"
|
|
hostReasons = $hostPerf.reasons
|
|
allowedReasons = $reasonMatches
|
|
}
|
|
continue
|
|
}
|
|
|
|
$cooldownMinutes = if ($action.cooldownMinutes) { [int]$action.cooldownMinutes } else { 30 }
|
|
$safeName = ($action.name -replace "[^A-Za-z0-9_.-]", "_")
|
|
$markerPath = Join-Path $EvidenceDir "loadshedding-$safeName.marker"
|
|
if (Test-Path $markerPath) {
|
|
$ageMinutes = ((Get-Date) - (Get-Item $markerPath).LastWriteTime).TotalMinutes
|
|
if ($ageMinutes -lt $cooldownMinutes) {
|
|
$results += [pscustomobject]@{ name = $action.name; host = $action.host; skipped = $true; reason = "cooldown"; ageMinutes = [math]::Round($ageMinutes, 1) }
|
|
continue
|
|
}
|
|
}
|
|
|
|
Record-AgentEvent "load_shed_attempt" $hostSeverity "host=$($action.host) action=$($action.name) command=$($action.command)" $action -NoAlert
|
|
$timeoutSeconds = if ($action.PSObject.Properties["timeoutSeconds"]) { [int]$action.timeoutSeconds } else { 60 }
|
|
$run = Invoke-HostSshText $action.host $action.command $timeoutSeconds 1
|
|
|
|
$postVerify = $null
|
|
$postVerifyOk = $false
|
|
if ($action.PSObject.Properties["postVerifyCommand"] -and $action.postVerifyCommand) {
|
|
$verifyRun = Invoke-HostSshText $action.host ([string]$action.postVerifyCommand) 30 1
|
|
$postVerify = [pscustomobject]@{
|
|
ok = [bool]$verifyRun.ok
|
|
exitCode = $verifyRun.exitCode
|
|
output = $verifyRun.output
|
|
route = $verifyRun.route
|
|
}
|
|
if ($verifyRun.ok) {
|
|
$postVerifyOk = $true
|
|
if ($action.PSObject.Properties["postVerifyMaxDiskUsedPercent"]) {
|
|
$match = [regex]::Match([string]$verifyRun.output, "__DISK_USED__=(\d+)")
|
|
if ($match.Success) {
|
|
$postVerifyOk = ([int]$match.Groups[1].Value -le [int]$action.postVerifyMaxDiskUsedPercent)
|
|
} else {
|
|
$postVerifyOk = $false
|
|
}
|
|
}
|
|
if ($action.PSObject.Properties["postVerifyOkRegex"] -and $action.postVerifyOkRegex) {
|
|
$postVerifyOk = [bool]([string]$verifyRun.output -match [string]$action.postVerifyOkRegex)
|
|
}
|
|
}
|
|
}
|
|
|
|
$effectiveOk = [bool]($run.ok -or $postVerifyOk)
|
|
Set-Content -Path $markerPath -Value (Get-Date -Format o) -Encoding UTF8
|
|
$result = [pscustomobject]@{
|
|
name = $action.name
|
|
host = $action.host
|
|
skipped = $false
|
|
ok = $effectiveOk
|
|
executionOk = [bool]$run.ok
|
|
postVerifyOk = [bool]$postVerifyOk
|
|
postVerify = $postVerify
|
|
route = $run.route
|
|
exitCode = $run.exitCode
|
|
output = $run.output
|
|
rollback = $action.rollback
|
|
}
|
|
Record-AgentEvent "load_shed_result" $(if ($effectiveOk) { "warning" } else { "critical" }) "host=$($action.host) action=$($action.name) ok=$effectiveOk executionOk=$($run.ok) postVerifyOk=$postVerifyOk" $result -Alert
|
|
$results += $result
|
|
}
|
|
|
|
$results
|
|
}
|
|
|
|
function Get-AgentBooleanSetting {
|
|
param(
|
|
[object]$Object,
|
|
[string]$Name,
|
|
[bool]$Default
|
|
)
|
|
|
|
if ($Object -and $Object.PSObject.Properties[$Name]) {
|
|
return [bool]$Object.PSObject.Properties[$Name].Value
|
|
}
|
|
return $Default
|
|
}
|
|
|
|
function Get-AgentSelfHealthConfig {
|
|
$selfHealth = $Config.selfHealth
|
|
$defaultTasks = @(
|
|
"Wooo-Agent99-Startup-Recovery",
|
|
"Wooo-Agent99-Heartbeat",
|
|
"Wooo-Agent99-Performance-Guard",
|
|
"Wooo-Agent99-Control-Loop",
|
|
"Wooo-Agent99-Telegram-Inbox",
|
|
"Wooo-Agent99-SRE-Alert-Inbox",
|
|
"Wooo-Agent99-Self-Health",
|
|
"Wooo-Agent99-Backup-Health"
|
|
)
|
|
$defaultEvidence = @(
|
|
[pscustomobject]@{ name = "performance_guard"; pattern = "agent99-Perf-*.json"; maxAgeMinutes = 5; required = $true },
|
|
[pscustomobject]@{ name = "heartbeat_status"; pattern = "agent99-Status-*.json"; maxAgeMinutes = 15; required = $true },
|
|
[pscustomobject]@{ name = "self_health"; pattern = "agent99-SelfCheck-*.json"; maxAgeMinutes = 15; required = $false },
|
|
[pscustomobject]@{ name = "backup_health"; pattern = "agent99-BackupCheck-*.json"; maxAgeMinutes = 30; required = $false },
|
|
[pscustomobject]@{ name = "startup_recovery"; pattern = "agent99-Recover-*.json"; maxAgeMinutes = 1440; required = $false },
|
|
[pscustomobject]@{ name = "telegram_inbox"; pattern = "agent99-TelegramInbox-*.json"; maxAgeMinutes = 15; required = $false },
|
|
[pscustomobject]@{ name = "sre_alert_inbox"; pattern = "agent99-SreAlertInbox-*.json"; maxAgeMinutes = 15; required = $false }
|
|
)
|
|
|
|
[pscustomobject]@{
|
|
scheduledTasks = if ($selfHealth -and $selfHealth.scheduledTasks) { @($selfHealth.scheduledTasks) } else { $defaultTasks }
|
|
evidenceFreshness = if ($selfHealth -and $selfHealth.evidenceFreshness) { @($selfHealth.evidenceFreshness) } else { $defaultEvidence }
|
|
requiredHosts = if ($selfHealth -and $selfHealth.requiredHosts) { @($selfHealth.requiredHosts) } else { @($Config.hosts) }
|
|
requireRecentTelegramDelivery = Get-AgentBooleanSetting $selfHealth "requireRecentTelegramDelivery" $true
|
|
}
|
|
}
|
|
|
|
function Get-AgentLatestEvidence {
|
|
param([string]$Pattern)
|
|
|
|
$file = Get-ChildItem -Path $EvidenceDir -Filter $Pattern -File -ErrorAction SilentlyContinue |
|
|
Sort-Object LastWriteTime -Descending |
|
|
Select-Object -First 1
|
|
if (-not $file) {
|
|
return [pscustomobject]@{
|
|
exists = $false
|
|
path = $null
|
|
fileName = $null
|
|
lastWriteTime = $null
|
|
ageMinutes = $null
|
|
data = $null
|
|
parseError = $null
|
|
}
|
|
}
|
|
|
|
$data = $null
|
|
$parseError = $null
|
|
try {
|
|
$data = Get-Content $file.FullName -Raw | ConvertFrom-Json
|
|
} catch {
|
|
$parseError = $_.Exception.Message
|
|
}
|
|
|
|
[pscustomobject]@{
|
|
exists = $true
|
|
path = $file.FullName
|
|
fileName = $file.Name
|
|
lastWriteTime = $file.LastWriteTime.ToString("o")
|
|
ageMinutes = [math]::Round(((Get-Date) - $file.LastWriteTime).TotalMinutes, 2)
|
|
data = $data
|
|
parseError = $parseError
|
|
}
|
|
}
|
|
|
|
function Test-AgentScheduledTaskHealth {
|
|
param([string[]]$TaskNames)
|
|
|
|
$results = @()
|
|
foreach ($taskName in $TaskNames) {
|
|
try {
|
|
$task = Get-ScheduledTask -TaskName $taskName -ErrorAction Stop
|
|
$info = Get-ScheduledTaskInfo -TaskName $taskName -ErrorAction Stop
|
|
$state = [string]$task.State
|
|
$lastTaskResult = $info.LastTaskResult
|
|
$pendingFirstRun = ($lastTaskResult -eq 267011)
|
|
$ok = ($state -eq "Running") -or (($state -eq "Ready") -and ($lastTaskResult -in @(0, 267011)))
|
|
$results += [pscustomobject]@{
|
|
name = $taskName
|
|
exists = $true
|
|
ok = [bool]$ok
|
|
state = $state
|
|
lastTaskResult = $lastTaskResult
|
|
pendingFirstRun = [bool]$pendingFirstRun
|
|
lastRunTime = if ($info.LastRunTime) { $info.LastRunTime.ToString("o") } else { $null }
|
|
nextRunTime = if ($info.NextRunTime) { $info.NextRunTime.ToString("o") } else { $null }
|
|
}
|
|
Write-AgentLog "selfcheck task name=$taskName ok=$ok state=$state lastTaskResult=$lastTaskResult pendingFirstRun=$pendingFirstRun"
|
|
} catch {
|
|
$results += [pscustomobject]@{
|
|
name = $taskName
|
|
exists = $false
|
|
ok = $false
|
|
state = $null
|
|
lastTaskResult = $null
|
|
lastRunTime = $null
|
|
nextRunTime = $null
|
|
error = $_.Exception.Message
|
|
}
|
|
Write-AgentLog "selfcheck task_missing name=$taskName error=$($_.Exception.Message)"
|
|
}
|
|
}
|
|
|
|
$results
|
|
}
|
|
|
|
function Test-AgentEvidenceFreshness {
|
|
param([object[]]$Contracts)
|
|
|
|
$results = @()
|
|
foreach ($contract in $Contracts) {
|
|
$name = if ($contract.name) { [string]$contract.name } else { [string]$contract.pattern }
|
|
$pattern = [string]$contract.pattern
|
|
$maxAgeMinutes = if ($contract.maxAgeMinutes) { [double]$contract.maxAgeMinutes } else { 15 }
|
|
$required = Get-AgentBooleanSetting $contract "required" $true
|
|
$latest = Get-AgentLatestEvidence $pattern
|
|
$fresh = [bool]($latest.exists -and $null -eq $latest.parseError -and $latest.ageMinutes -le $maxAgeMinutes)
|
|
$ok = [bool](((-not $required) -and (-not $latest.exists)) -or $fresh)
|
|
$severity = if ($ok) { "ok" } elseif ($required) { "critical" } else { "warning" }
|
|
|
|
$results += [pscustomobject]@{
|
|
name = $name
|
|
pattern = $pattern
|
|
required = [bool]$required
|
|
ok = $ok
|
|
severity = $severity
|
|
latestPath = $latest.path
|
|
latestFile = $latest.fileName
|
|
lastWriteTime = $latest.lastWriteTime
|
|
ageMinutes = $latest.ageMinutes
|
|
maxAgeMinutes = $maxAgeMinutes
|
|
parseError = $latest.parseError
|
|
}
|
|
Write-AgentLog "selfcheck evidence name=$name ok=$ok severity=$severity ageMinutes=$($latest.ageMinutes) maxAgeMinutes=$maxAgeMinutes file=$($latest.fileName)"
|
|
}
|
|
|
|
$results
|
|
}
|
|
|
|
function Test-AgentLatestPerformanceEvidence {
|
|
param([string[]]$RequiredHosts)
|
|
|
|
$latest = Get-AgentLatestEvidence "agent99-Perf-*.json"
|
|
$hostChecks = @()
|
|
if (-not $latest.exists) {
|
|
return [pscustomobject]@{
|
|
ok = $false
|
|
severity = "critical"
|
|
reason = "missing_perf_evidence"
|
|
latestPath = $null
|
|
latestFile = $null
|
|
ageMinutes = $null
|
|
hosts = $hostChecks
|
|
}
|
|
}
|
|
if ($latest.parseError) {
|
|
return [pscustomobject]@{
|
|
ok = $false
|
|
severity = "critical"
|
|
reason = "perf_evidence_parse_error"
|
|
latestPath = $latest.path
|
|
latestFile = $latest.fileName
|
|
ageMinutes = $latest.ageMinutes
|
|
parseError = $latest.parseError
|
|
hosts = $hostChecks
|
|
}
|
|
}
|
|
|
|
$perf = @($latest.data.performance)
|
|
foreach ($hostIp in $RequiredHosts) {
|
|
$entry = $perf | Where-Object { $_.host -eq $hostIp } | Select-Object -First 1
|
|
$hostSeverity = if (-not $entry) {
|
|
"missing"
|
|
} elseif ($entry.severity) {
|
|
[string]$entry.severity
|
|
} elseif ($entry.ok -eq $false) {
|
|
"warning"
|
|
} else {
|
|
"ok"
|
|
}
|
|
$ok = [bool]($hostSeverity -eq "ok")
|
|
$hostChecks += [pscustomobject]@{
|
|
host = $hostIp
|
|
present = [bool]$entry
|
|
ok = $ok
|
|
severity = $hostSeverity
|
|
loadPerCore = if ($entry) { $entry.loadPerCore } else { $null }
|
|
memAvailablePercent = if ($entry) { $entry.memAvailablePercent } else { $null }
|
|
diskUsedPercent = if ($entry) { $entry.diskUsedPercent } else { $null }
|
|
reasons = if ($entry) { @($entry.reasons) } else { @("missing_perf_host") }
|
|
}
|
|
}
|
|
|
|
$missingOrCritical = @($hostChecks | Where-Object { $_.severity -in @("missing", "critical") }).Count
|
|
$warning = @($hostChecks | Where-Object { $_.severity -eq "warning" }).Count
|
|
$severity = if ($missingOrCritical -gt 0) { "critical" } elseif ($warning -gt 0) { "warning" } else { "ok" }
|
|
|
|
[pscustomobject]@{
|
|
ok = [bool]($severity -eq "ok")
|
|
severity = $severity
|
|
reason = if ($severity -eq "ok") { "all_required_hosts_ok" } else { "required_host_perf_not_ok" }
|
|
latestPath = $latest.path
|
|
latestFile = $latest.fileName
|
|
ageMinutes = $latest.ageMinutes
|
|
hosts = $hostChecks
|
|
}
|
|
}
|
|
|
|
function Test-AgentRecentTelegramDelivery {
|
|
param([object[]]$Contracts)
|
|
|
|
$checks = @()
|
|
$lookbackMinutes = 60
|
|
if ($Config.selfHealth -and $Config.selfHealth.PSObject.Properties["telegramDeliveryLookbackMinutes"]) {
|
|
$lookbackMinutes = [int]$Config.selfHealth.telegramDeliveryLookbackMinutes
|
|
}
|
|
$cutoff = (Get-Date).AddMinutes(-1 * $lookbackMinutes)
|
|
foreach ($contract in $Contracts) {
|
|
$pattern = [string]$contract.pattern
|
|
$files = Get-ChildItem -Path $EvidenceDir -Filter $pattern -File -ErrorAction SilentlyContinue |
|
|
Where-Object { $_.LastWriteTime -ge $cutoff } |
|
|
Sort-Object LastWriteTime -Descending |
|
|
Select-Object -First 20
|
|
|
|
foreach ($file in @($files)) {
|
|
$data = $null
|
|
try {
|
|
$data = Get-Content $file.FullName -Raw | ConvertFrom-Json
|
|
} catch {
|
|
continue
|
|
}
|
|
if (-not $data) { continue }
|
|
$attempts = @($data.telegram)
|
|
if ($attempts.Count -eq 0) { continue }
|
|
|
|
$sent = @($attempts | Where-Object { $_.sent -eq $true -or ($_.relay -and $_.relay.ok -eq $true) })
|
|
$checks += [pscustomobject]@{
|
|
name = if ($contract.name) { [string]$contract.name } else { [string]$contract.pattern }
|
|
latestPath = $file.FullName
|
|
latestFile = $file.Name
|
|
ageMinutes = [math]::Round(((Get-Date) - $file.LastWriteTime).TotalMinutes, 2)
|
|
attempts = $attempts.Count
|
|
sent = $sent.Count
|
|
ok = [bool]($sent.Count -gt 0)
|
|
}
|
|
}
|
|
}
|
|
|
|
$attemptCount = 0
|
|
$sentCount = 0
|
|
foreach ($check in @($checks)) {
|
|
$attemptCount += [int]$check.attempts
|
|
$sentCount += [int]$check.sent
|
|
}
|
|
if ($attemptCount -eq 0) {
|
|
return [pscustomobject]@{
|
|
ok = $true
|
|
severity = "ok"
|
|
reason = "no_recent_alert_attempts"
|
|
checks = $checks
|
|
}
|
|
}
|
|
|
|
[pscustomobject]@{
|
|
ok = [bool]($sentCount -gt 0)
|
|
severity = if ($sentCount -gt 0) { "ok" } else { "critical" }
|
|
reason = if ($sentCount -gt 0) { "recent_alert_delivery_ok" } else { "recent_alert_delivery_failed" }
|
|
checks = $checks
|
|
}
|
|
}
|
|
|
|
function Test-AgentSelfHealth {
|
|
$selfConfig = Get-AgentSelfHealthConfig
|
|
$tasks = Test-AgentScheduledTaskHealth $selfConfig.scheduledTasks
|
|
$evidence = Test-AgentEvidenceFreshness $selfConfig.evidenceFreshness
|
|
$latestPerf = Test-AgentLatestPerformanceEvidence $selfConfig.requiredHosts
|
|
$telegram = if ($selfConfig.requireRecentTelegramDelivery) {
|
|
Test-AgentRecentTelegramDelivery $selfConfig.evidenceFreshness
|
|
} else {
|
|
[pscustomobject]@{ ok = $true; severity = "ok"; checks = @(); skipped = $true }
|
|
}
|
|
|
|
$taskFailures = @($tasks | Where-Object { -not $_.ok }).Count
|
|
$evidenceCritical = @($evidence | Where-Object { $_.severity -eq "critical" }).Count
|
|
$evidenceWarning = @($evidence | Where-Object { $_.severity -eq "warning" }).Count
|
|
$critical = $taskFailures + $evidenceCritical
|
|
if ($latestPerf.severity -eq "critical") { $critical += 1 }
|
|
if ($telegram.severity -eq "critical") { $critical += 1 }
|
|
$warning = $evidenceWarning
|
|
if ($latestPerf.severity -eq "warning") { $warning += 1 }
|
|
|
|
$severity = if ($critical -gt 0) { "critical" } elseif ($warning -gt 0) { "warning" } else { "ok" }
|
|
$message = "selfHealth severity=$severity taskFailures=$taskFailures evidenceCritical=$evidenceCritical evidenceWarning=$evidenceWarning perf=$($latestPerf.severity) telegram=$($telegram.severity)"
|
|
Record-AgentEvent "agent_selfcheck" $(if ($severity -eq "ok") { "info" } else { $severity }) $message $null -Alert
|
|
|
|
[pscustomobject]@{
|
|
ok = [bool]($severity -eq "ok")
|
|
severity = $severity
|
|
tasks = $tasks
|
|
evidenceFreshness = $evidence
|
|
latestPerformance = $latestPerf
|
|
telegramDelivery = $telegram
|
|
summary = [pscustomobject]@{
|
|
taskFailures = $taskFailures
|
|
evidenceCritical = $evidenceCritical
|
|
evidenceWarning = $evidenceWarning
|
|
performanceSeverity = $latestPerf.severity
|
|
telegramSeverity = $telegram.severity
|
|
}
|
|
}
|
|
}
|
|
|
|
function Get-AgentUnixAgeMinutes {
|
|
param([double]$UnixSeconds)
|
|
|
|
$now = [DateTimeOffset]::Now.ToUnixTimeSeconds()
|
|
[math]::Round(($now - $UnixSeconds) / 60, 2)
|
|
}
|
|
|
|
function Convert-AgentUnixToIso {
|
|
param([double]$UnixSeconds)
|
|
|
|
try {
|
|
[DateTimeOffset]::FromUnixTimeSeconds([int64][math]::Floor($UnixSeconds)).LocalDateTime.ToString("o")
|
|
} catch {
|
|
$null
|
|
}
|
|
}
|
|
|
|
function Get-AgentBackupHealthConfig {
|
|
$backupHealth = $Config.backupHealth
|
|
$backupHost = if ($backupHealth -and $backupHealth.host) { [string]$backupHealth.host } else { "192.168.0.110" }
|
|
$targets = @(
|
|
[pscustomobject]@{ name = "awoooi"; path = "/backup/awoooi/snapshots"; maxAgeMinutes = 430; required = $true },
|
|
[pscustomobject]@{ name = "gitea"; path = "/backup/gitea/snapshots"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "harbor"; path = "/backup/harbor/snapshots"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "momo"; path = "/backup/momo/snapshots"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "langfuse"; path = "/backup/langfuse/snapshots"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "monitoring"; path = "/backup/monitoring/snapshots"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "signoz"; path = "/backup/signoz/snapshots"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "clawbot"; path = "/backup/clawbot/snapshots"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "open-webui"; path = "/backup/open-webui/snapshots"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "sentry"; path = "/backup/sentry/snapshots"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "ai-artifacts"; path = "/backup/ai-artifacts/snapshots"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "configs"; path = "/backup/configs/snapshots"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "public-routes"; path = "/backup/public-routes/snapshots"; maxAgeMinutes = 10080; required = $false }
|
|
)
|
|
$fileChecks = @(
|
|
[pscustomobject]@{ name = "backup_log"; path = "/backup/logs/backup.log"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "backup_status_log"; path = "/backup/logs/backup-status.log"; maxAgeMinutes = 1800; required = $true },
|
|
[pscustomobject]@{ name = "status_log"; path = "/backup/logs/status.log"; maxAgeMinutes = 10080; required = $false },
|
|
[pscustomobject]@{ name = "cron_log"; path = "/backup/logs/cron.log"; maxAgeMinutes = 10080; required = $false },
|
|
[pscustomobject]@{ name = "integrity_check_status"; path = "/backup/integrity/check.status"; maxAgeMinutes = 10080; required = $true },
|
|
[pscustomobject]@{ name = "restore_drill_status"; path = "/backup/integrity/restore-drill.status"; maxAgeMinutes = 45000; required = $false },
|
|
[pscustomobject]@{ name = "configs_last_status"; path = "/backup/status/backup-configs-last-status.json"; maxAgeMinutes = 1800; required = $false }
|
|
)
|
|
$cronPatterns = @(
|
|
"backup-all.sh",
|
|
"backup-awoooi-frequent.sh",
|
|
"backup-health-textfile-exporter.py",
|
|
"check-backup-integrity.sh",
|
|
"sync-offsite-backups.sh --mode status",
|
|
"backup-status.sh"
|
|
)
|
|
|
|
[pscustomobject]@{
|
|
host = $backupHost
|
|
targets = if ($backupHealth -and $backupHealth.targets) { @($backupHealth.targets) } else { $targets }
|
|
fileChecks = if ($backupHealth -and $backupHealth.fileChecks) { @($backupHealth.fileChecks) } else { $fileChecks }
|
|
cronPatterns = if ($backupHealth -and $backupHealth.cronPatterns) { @($backupHealth.cronPatterns) } else { $cronPatterns }
|
|
}
|
|
}
|
|
|
|
function Test-AgentBackupSnapshotTarget {
|
|
param(
|
|
[object]$Target,
|
|
[string]$DefaultHost
|
|
)
|
|
|
|
$name = [string]$Target.name
|
|
$hostIp = if ($Target.host) { [string]$Target.host } else { $DefaultHost }
|
|
$path = [string]$Target.path
|
|
$required = Get-AgentBooleanSetting $Target "required" $true
|
|
$maxAgeMinutes = if ($Target.maxAgeMinutes) { [double]$Target.maxAgeMinutes } else { 1800 }
|
|
|
|
if ($path -notmatch "^/backup/[A-Za-z0-9_.\/-]+$") {
|
|
return [pscustomobject]@{ name = $name; host = $hostIp; path = $path; ok = $false; severity = "critical"; reason = "invalid_backup_path" }
|
|
}
|
|
|
|
$quotedPath = Quote-ShSingle $path
|
|
$command = "if [ -d $quotedPath ]; then echo __COUNT__=`$(find $quotedPath -maxdepth 1 -type f 2>/dev/null | wc -l); find $quotedPath -maxdepth 1 -type f -printf '__LATEST__=%T@|%TY-%Tm-%TdT%TH:%TM:%TS|%s|%p\n' 2>/dev/null | sort -n | tail -1; else echo __MISSING_DIR__=$quotedPath; fi"
|
|
$timeoutSeconds = 60
|
|
if ($Config.backupHealth -and $Config.backupHealth.PSObject.Properties["snapshotReadTimeoutSeconds"]) {
|
|
$timeoutSeconds = [int]$Config.backupHealth.snapshotReadTimeoutSeconds
|
|
}
|
|
if ($Target.PSObject.Properties["timeoutSeconds"]) {
|
|
$timeoutSeconds = [int]$Target.timeoutSeconds
|
|
}
|
|
$readback = Invoke-HostSshText $hostIp $command $timeoutSeconds 1
|
|
|
|
$count = 0
|
|
$latestUnix = $null
|
|
$latestSize = $null
|
|
$latestPath = $null
|
|
if ($readback.ok) {
|
|
$countMatch = [regex]::Match($readback.output, "__COUNT__=(\d+)")
|
|
if ($countMatch.Success) { $count = [int]$countMatch.Groups[1].Value }
|
|
$latestMatch = [regex]::Match($readback.output, "__LATEST__=([0-9.]+)\|([^|]+)\|(\d+)\|(.+)")
|
|
if ($latestMatch.Success) {
|
|
$latestUnix = [double]::Parse($latestMatch.Groups[1].Value, [Globalization.CultureInfo]::InvariantCulture)
|
|
$latestSize = [int64]$latestMatch.Groups[3].Value
|
|
$latestPath = $latestMatch.Groups[4].Value.Trim()
|
|
}
|
|
}
|
|
|
|
$ageMinutes = if ($null -ne $latestUnix) { Get-AgentUnixAgeMinutes $latestUnix } else { $null }
|
|
$severity = "ok"
|
|
$reason = "fresh"
|
|
if (-not $readback.ok) {
|
|
$severity = if ($required) { "critical" } else { "warning" }
|
|
$reason = "backup_readback_failed"
|
|
} elseif ($readback.output -match "__MISSING_DIR__" -or $count -eq 0 -or $null -eq $latestUnix) {
|
|
$severity = if ($required) { "critical" } else { "warning" }
|
|
$reason = "backup_snapshot_missing"
|
|
} elseif ($ageMinutes -gt $maxAgeMinutes) {
|
|
$severity = if ($required) { "critical" } else { "warning" }
|
|
$reason = "backup_snapshot_stale"
|
|
}
|
|
|
|
$result = [pscustomobject]@{
|
|
name = $name
|
|
host = $hostIp
|
|
path = $path
|
|
required = [bool]$required
|
|
ok = [bool]($severity -eq "ok")
|
|
severity = $severity
|
|
reason = $reason
|
|
count = $count
|
|
latestPath = $latestPath
|
|
latestSizeBytes = $latestSize
|
|
latestTime = if ($null -ne $latestUnix) { Convert-AgentUnixToIso $latestUnix } else { $null }
|
|
ageMinutes = $ageMinutes
|
|
maxAgeMinutes = $maxAgeMinutes
|
|
route = $readback.route
|
|
exitCode = $readback.exitCode
|
|
}
|
|
Write-AgentLog "backup target=$name severity=$severity reason=$reason ageMinutes=$ageMinutes count=$count"
|
|
if ($severity -in @("warning", "critical")) {
|
|
Record-AgentEvent "backup_$reason" $severity "target=$name host=$hostIp ageMinutes=$ageMinutes maxAgeMinutes=$maxAgeMinutes path=$path" $result -Alert
|
|
}
|
|
$result
|
|
}
|
|
|
|
function Test-AgentBackupFileCheck {
|
|
param(
|
|
[object]$FileCheck,
|
|
[string]$DefaultHost
|
|
)
|
|
|
|
$name = [string]$FileCheck.name
|
|
$hostIp = if ($FileCheck.host) { [string]$FileCheck.host } else { $DefaultHost }
|
|
$path = [string]$FileCheck.path
|
|
$required = Get-AgentBooleanSetting $FileCheck "required" $true
|
|
$maxAgeMinutes = if ($FileCheck.maxAgeMinutes) { [double]$FileCheck.maxAgeMinutes } else { 1800 }
|
|
|
|
if ($path -notmatch "^/backup/[A-Za-z0-9_.\/-]+$") {
|
|
return [pscustomobject]@{ name = $name; host = $hostIp; path = $path; ok = $false; severity = "critical"; reason = "invalid_backup_file_path" }
|
|
}
|
|
|
|
$quotedPath = Quote-ShSingle $path
|
|
$command = "if [ -f $quotedPath ]; then stat -c '__STAT__=%Y|%y|%s|%n' $quotedPath; else echo __MISSING_FILE__=$quotedPath; fi"
|
|
$readback = Invoke-HostSshText $hostIp $command 15 1
|
|
$unix = $null
|
|
$size = $null
|
|
if ($readback.ok) {
|
|
$statMatch = [regex]::Match($readback.output, "__STAT__=([0-9]+)\|([^|]+)\|(\d+)\|(.+)")
|
|
if ($statMatch.Success) {
|
|
$unix = [double]$statMatch.Groups[1].Value
|
|
$size = [int64]$statMatch.Groups[3].Value
|
|
}
|
|
}
|
|
|
|
$ageMinutes = if ($null -ne $unix) { Get-AgentUnixAgeMinutes $unix } else { $null }
|
|
$severity = "ok"
|
|
$reason = "fresh"
|
|
if (-not $readback.ok) {
|
|
$severity = if ($required) { "critical" } else { "warning" }
|
|
$reason = "backup_file_readback_failed"
|
|
} elseif ($readback.output -match "__MISSING_FILE__" -or $null -eq $unix) {
|
|
$severity = if ($required) { "critical" } else { "warning" }
|
|
$reason = "backup_file_missing"
|
|
} elseif ($ageMinutes -gt $maxAgeMinutes) {
|
|
$severity = if ($required) { "critical" } else { "warning" }
|
|
$reason = "backup_file_stale"
|
|
}
|
|
|
|
$result = [pscustomobject]@{
|
|
name = $name
|
|
host = $hostIp
|
|
path = $path
|
|
required = [bool]$required
|
|
ok = [bool]($severity -eq "ok")
|
|
severity = $severity
|
|
reason = $reason
|
|
sizeBytes = $size
|
|
lastWriteTime = if ($null -ne $unix) { Convert-AgentUnixToIso $unix } else { $null }
|
|
ageMinutes = $ageMinutes
|
|
maxAgeMinutes = $maxAgeMinutes
|
|
route = $readback.route
|
|
exitCode = $readback.exitCode
|
|
}
|
|
Write-AgentLog "backup file=$name severity=$severity reason=$reason ageMinutes=$ageMinutes"
|
|
if ($severity -in @("warning", "critical")) {
|
|
Record-AgentEvent "backup_$reason" $severity "file=$name host=$hostIp ageMinutes=$ageMinutes maxAgeMinutes=$maxAgeMinutes path=$path" $result -Alert
|
|
}
|
|
$result
|
|
}
|
|
|
|
function Test-AgentBackupCronPattern {
|
|
param(
|
|
[string]$Pattern,
|
|
[string]$DefaultHost
|
|
)
|
|
|
|
$quotedPattern = Quote-ShSingle $Pattern
|
|
$command = "crontab -l 2>/dev/null | grep -F -- $quotedPattern >/dev/null && echo __PRESENT__ || echo __MISSING__"
|
|
$readback = Invoke-HostSshText $DefaultHost $command 8 2
|
|
$present = [bool]($readback.ok -and $readback.output -match "__PRESENT__")
|
|
$result = [pscustomobject]@{
|
|
host = $DefaultHost
|
|
pattern = $Pattern
|
|
ok = $present
|
|
severity = if ($present) { "ok" } else { "critical" }
|
|
reason = if ($present) { "cron_pattern_present" } else { "cron_pattern_missing" }
|
|
route = $readback.route
|
|
exitCode = $readback.exitCode
|
|
}
|
|
Write-AgentLog "backup cron pattern=$Pattern ok=$present"
|
|
if (-not $present) {
|
|
Record-AgentEvent "backup_cron_pattern_missing" "critical" "host=$DefaultHost pattern=$Pattern" $result -Alert
|
|
}
|
|
$result
|
|
}
|
|
|
|
function Test-AgentBackupCronPatterns {
|
|
param(
|
|
[string[]]$Patterns,
|
|
[string]$DefaultHost
|
|
)
|
|
|
|
$readback = Invoke-HostSshText $DefaultHost "crontab -l 2>/dev/null" 12 2
|
|
$results = @()
|
|
if (-not $readback.ok) {
|
|
foreach ($pattern in $Patterns) {
|
|
$result = [pscustomobject]@{
|
|
host = $DefaultHost
|
|
pattern = $pattern
|
|
ok = $false
|
|
severity = "warning"
|
|
reason = "cron_readback_failed"
|
|
route = $readback.route
|
|
exitCode = $readback.exitCode
|
|
}
|
|
Record-AgentEvent "backup_cron_readback_failed" "warning" "host=$DefaultHost pattern=$pattern" $result -Alert
|
|
$results += $result
|
|
}
|
|
return $results
|
|
}
|
|
|
|
$crontab = [string]$readback.output
|
|
foreach ($pattern in $Patterns) {
|
|
$present = [bool]($crontab -match [regex]::Escape($pattern))
|
|
$result = [pscustomobject]@{
|
|
host = $DefaultHost
|
|
pattern = $pattern
|
|
ok = $present
|
|
severity = if ($present) { "ok" } else { "critical" }
|
|
reason = if ($present) { "cron_pattern_present" } else { "cron_pattern_missing" }
|
|
route = $readback.route
|
|
exitCode = $readback.exitCode
|
|
}
|
|
Write-AgentLog "backup cron pattern=$pattern ok=$present"
|
|
if (-not $present) {
|
|
Record-AgentEvent "backup_cron_pattern_missing" "critical" "host=$DefaultHost pattern=$pattern" $result -Alert
|
|
}
|
|
$results += $result
|
|
}
|
|
|
|
$results
|
|
}
|
|
|
|
function Test-AgentBackupHealth {
|
|
$backupConfig = Get-AgentBackupHealthConfig
|
|
$targets = @()
|
|
foreach ($target in @($backupConfig.targets)) {
|
|
$targets += Test-AgentBackupSnapshotTarget $target $backupConfig.host
|
|
}
|
|
|
|
$files = @()
|
|
foreach ($fileCheck in @($backupConfig.fileChecks)) {
|
|
$files += Test-AgentBackupFileCheck $fileCheck $backupConfig.host
|
|
}
|
|
|
|
$cron = Test-AgentBackupCronPatterns @($backupConfig.cronPatterns) $backupConfig.host
|
|
|
|
$targetCritical = @($targets | Where-Object { $_.severity -eq "critical" }).Count
|
|
$targetWarning = @($targets | Where-Object { $_.severity -eq "warning" }).Count
|
|
$fileCritical = @($files | Where-Object { $_.severity -eq "critical" }).Count
|
|
$fileWarning = @($files | Where-Object { $_.severity -eq "warning" }).Count
|
|
$cronCritical = @($cron | Where-Object { $_.severity -eq "critical" }).Count
|
|
$cronWarning = @($cron | Where-Object { $_.severity -eq "warning" }).Count
|
|
$critical = $targetCritical + $fileCritical + $cronCritical
|
|
$warning = $targetWarning + $fileWarning + $cronWarning
|
|
$severity = if ($critical -gt 0) { "critical" } elseif ($warning -gt 0) { "warning" } else { "ok" }
|
|
|
|
$summary = [pscustomobject]@{
|
|
targetCritical = $targetCritical
|
|
targetWarning = $targetWarning
|
|
fileCritical = $fileCritical
|
|
fileWarning = $fileWarning
|
|
cronCritical = $cronCritical
|
|
cronWarning = $cronWarning
|
|
targetCount = @($targets).Count
|
|
fileCheckCount = @($files).Count
|
|
cronPatternCount = @($cron).Count
|
|
}
|
|
if ($severity -in @("warning", "critical")) {
|
|
Record-AgentEvent "backup_health_degraded" $severity "severity=$severity targetCritical=$targetCritical fileCritical=$fileCritical cronCritical=$cronCritical" $summary -Alert
|
|
} else {
|
|
Write-AgentLog "backup_health ok targets=$($summary.targetCount) files=$($summary.fileCheckCount) cron=$($summary.cronPatternCount)"
|
|
}
|
|
|
|
[pscustomobject]@{
|
|
ok = [bool]($severity -eq "ok")
|
|
severity = $severity
|
|
host = $backupConfig.host
|
|
targets = $targets
|
|
files = $files
|
|
cron = $cron
|
|
summary = $summary
|
|
}
|
|
}
|
|
|
|
function Convert-AgentHtml {
|
|
param([object]$Value)
|
|
[System.Net.WebUtility]::HtmlEncode([string]$Value)
|
|
}
|
|
|
|
function Get-AgentLatestEvidenceSummary {
|
|
param(
|
|
[string]$Name,
|
|
[string]$Pattern
|
|
)
|
|
|
|
$latest = Get-AgentLatestEvidence $Pattern
|
|
[pscustomobject]@{
|
|
name = $Name
|
|
exists = $latest.exists
|
|
file = $latest.fileName
|
|
path = $latest.path
|
|
ageMinutes = $latest.ageMinutes
|
|
parseError = $latest.parseError
|
|
}
|
|
}
|
|
|
|
function Convert-AgentSafeKey {
|
|
param([string]$Value)
|
|
$safe = [regex]::Replace(([string]$Value), "[^A-Za-z0-9_.:-]", "_")
|
|
if (-not $safe) { return "unknown" }
|
|
if ($safe.Length -gt 140) { return $safe.Substring(0, 140) }
|
|
return $safe
|
|
}
|
|
|
|
function Get-AgentProblemStats {
|
|
$stateDir = Join-Path $Config.agentRoot "state"
|
|
$statsPath = Join-Path $stateDir "problem-counts.json"
|
|
if (-not (Test-Path $statsPath)) {
|
|
return [pscustomobject]@{
|
|
path = $statsPath
|
|
totalEvents = 0
|
|
totalResolved = 0
|
|
totalFailed = 0
|
|
problems = @()
|
|
}
|
|
}
|
|
try {
|
|
$data = Get-Content $statsPath -Raw | ConvertFrom-Json
|
|
return [pscustomobject]@{
|
|
path = $statsPath
|
|
totalEvents = if ($data.totalEvents) { [int]$data.totalEvents } else { 0 }
|
|
totalResolved = if ($data.totalResolved) { [int]$data.totalResolved } else { 0 }
|
|
totalFailed = if ($data.totalFailed) { [int]$data.totalFailed } else { 0 }
|
|
problems = @($data.problems)
|
|
}
|
|
} catch {
|
|
Write-AgentLog "problem_stats_parse_failed path=$statsPath error=$($_.Exception.Message)"
|
|
return [pscustomobject]@{
|
|
path = $statsPath
|
|
totalEvents = 0
|
|
totalResolved = 0
|
|
totalFailed = 0
|
|
problems = @()
|
|
parseError = $_.Exception.Message
|
|
}
|
|
}
|
|
}
|
|
|
|
function Update-AgentProblemKnowledge {
|
|
param([object]$Result)
|
|
|
|
$stateDir = Join-Path $Config.agentRoot "state"
|
|
$playbookDir = Join-Path $Config.agentRoot "playbooks"
|
|
$ragDir = Join-Path $playbookDir "rag-import"
|
|
New-Item -ItemType Directory -Force -Path $stateDir, $playbookDir, $ragDir | Out-Null
|
|
|
|
$statsPath = Join-Path $stateDir "problem-counts.json"
|
|
$eventsPath = Join-Path $playbookDir "agent99-incident-records.jsonl"
|
|
$kmPath = Join-Path $playbookDir "agent99-incident-km.md"
|
|
$ragPath = Join-Path $ragDir "agent99-incidents.jsonl"
|
|
|
|
$source = if ($Result.source) { [string]$Result.source } else { "agent99" }
|
|
$kind = if ($Result.alertKind) { [string]$Result.alertKind } elseif ($Result.reason) { [string]$Result.reason } else { [string]$Result.mode }
|
|
$service = if ($Result.alertService) { [string]$Result.alertService } else { "agent99" }
|
|
$hostName = if ($Result.alertHost) { [string]$Result.alertHost } else { "" }
|
|
$severity = if ($Result.alertSeverity) { [string]$Result.alertSeverity } elseif ($Result.ok) { "info" } else { "critical" }
|
|
$key = Convert-AgentSafeKey (($source, $service, $hostName, $kind | Where-Object { $_ }) -join "|")
|
|
$now = Get-Date -Format o
|
|
$statusText = if ($Result.ok) { "resolved" } else { "failed" }
|
|
|
|
$event = [pscustomobject]@{
|
|
timestamp = $now
|
|
problemKey = $key
|
|
status = $statusText
|
|
commandId = $Result.id
|
|
source = $source
|
|
reason = $Result.reason
|
|
alertId = $Result.alertId
|
|
alertKind = $kind
|
|
alertSeverity = $severity
|
|
service = $service
|
|
host = $hostName
|
|
mode = $Result.mode
|
|
instruction = $Result.instruction
|
|
ok = [bool]$Result.ok
|
|
exitCode = $Result.exitCode
|
|
evidence = $Result.evidence
|
|
}
|
|
|
|
($event | ConvertTo-Json -Depth 8 -Compress) | Add-Content -Path $eventsPath -Encoding UTF8
|
|
($event | ConvertTo-Json -Depth 8 -Compress) | Add-Content -Path $ragPath -Encoding UTF8
|
|
|
|
$stats = Get-AgentProblemStats
|
|
$problems = @($stats.problems)
|
|
$existing = $problems | Where-Object { $_.key -eq $key } | Select-Object -First 1
|
|
if (-not $existing) {
|
|
$existing = [pscustomobject]@{
|
|
key = $key
|
|
source = $source
|
|
kind = $kind
|
|
service = $service
|
|
host = $hostName
|
|
severity = $severity
|
|
occurrences = 0
|
|
resolved = 0
|
|
failed = 0
|
|
lastStatus = $null
|
|
firstSeenAt = $now
|
|
lastSeenAt = $null
|
|
lastResolvedAt = $null
|
|
lastEvidence = $null
|
|
lastMode = $null
|
|
lastInstruction = $null
|
|
}
|
|
$problems += $existing
|
|
}
|
|
|
|
$existing.occurrences = [int]$existing.occurrences + 1
|
|
if ($Result.ok) {
|
|
$existing.resolved = [int]$existing.resolved + 1
|
|
$existing.lastResolvedAt = $now
|
|
} else {
|
|
$existing.failed = [int]$existing.failed + 1
|
|
}
|
|
$existing.lastStatus = $statusText
|
|
$existing.lastSeenAt = $now
|
|
$existing.lastEvidence = $Result.evidence
|
|
$existing.lastMode = $Result.mode
|
|
$existing.lastInstruction = $Result.instruction
|
|
$existing.severity = $severity
|
|
|
|
$newTotalEvents = [int]$stats.totalEvents + 1
|
|
$newTotalResolved = [int]$stats.totalResolved + $(if ($Result.ok) { 1 } else { 0 })
|
|
$newTotalFailed = [int]$stats.totalFailed + $(if ($Result.ok) { 0 } else { 1 })
|
|
|
|
$updated = [pscustomobject]@{
|
|
generatedAt = $now
|
|
totalEvents = $newTotalEvents
|
|
totalResolved = $newTotalResolved
|
|
totalFailed = $newTotalFailed
|
|
problems = @($problems | Sort-Object @{ Expression = { [int]$_.occurrences }; Descending = $true }, @{ Expression = { $_.lastSeenAt }; Descending = $true })
|
|
}
|
|
$updated | ConvertTo-Json -Depth 10 | Set-Content -Path $statsPath -Encoding UTF8
|
|
|
|
$rows = @($updated.problems | Select-Object -First 50 | ForEach-Object {
|
|
"- key=$($_.key); occurrences=$($_.occurrences); resolved=$($_.resolved); failed=$($_.failed); lastStatus=$($_.lastStatus); lastMode=$($_.lastMode); evidence=$($_.lastEvidence)"
|
|
})
|
|
$km = @(
|
|
"# Agent99 Incident KM",
|
|
"",
|
|
"Last updated: $now",
|
|
"",
|
|
"## Totals",
|
|
"",
|
|
"- totalEvents: $($updated.totalEvents)",
|
|
"- totalResolved: $($updated.totalResolved)",
|
|
"- totalFailed: $($updated.totalFailed)",
|
|
"",
|
|
"## Recurring Problems",
|
|
"",
|
|
($rows -join "`n")
|
|
) -join "`n"
|
|
Set-Content -Path $kmPath -Value $km -Encoding UTF8
|
|
|
|
[pscustomobject]@{
|
|
key = $key
|
|
occurrences = $existing.occurrences
|
|
resolved = $existing.resolved
|
|
failed = $existing.failed
|
|
lastStatus = $existing.lastStatus
|
|
statsPath = $statsPath
|
|
eventsPath = $eventsPath
|
|
kmPath = $kmPath
|
|
ragPath = $ragPath
|
|
}
|
|
}
|
|
|
|
function Get-AgentDashboardSummary {
|
|
$status = Get-AgentLatestEvidence "agent99-Status-*.json"
|
|
$perf = Get-AgentLatestEvidence "agent99-Perf-*.json"
|
|
$self = Get-AgentLatestEvidence "agent99-SelfCheck-*.json"
|
|
$backup = Get-AgentLatestEvidence "agent99-BackupCheck-*.json"
|
|
$recover = Get-AgentLatestEvidence "agent99-Recover-*.json"
|
|
|
|
$taskNames = @(
|
|
"Wooo-Agent99-Startup-Recovery",
|
|
"Wooo-Agent99-Heartbeat",
|
|
"Wooo-Agent99-Performance-Guard",
|
|
"Wooo-Agent99-Control-Loop",
|
|
"Wooo-Agent99-Telegram-Inbox",
|
|
"Wooo-Agent99-SRE-Alert-Inbox",
|
|
"Wooo-Agent99-Self-Health",
|
|
"Wooo-Agent99-Backup-Health"
|
|
)
|
|
$tasks = @()
|
|
foreach ($taskName in $taskNames) {
|
|
try {
|
|
$task = Get-ScheduledTask -TaskName $taskName -ErrorAction Stop
|
|
$info = Get-ScheduledTaskInfo -TaskName $taskName -ErrorAction Stop
|
|
$tasks += [pscustomobject]@{
|
|
name = $taskName
|
|
exists = $true
|
|
state = [string]$task.State
|
|
lastTaskResult = $info.LastTaskResult
|
|
lastRunTime = if ($info.LastRunTime) { $info.LastRunTime.ToString("yyyy-MM-dd HH:mm:ss") } else { $null }
|
|
nextRunTime = if ($info.NextRunTime) { $info.NextRunTime.ToString("yyyy-MM-dd HH:mm:ss") } else { $null }
|
|
}
|
|
} catch {
|
|
$tasks += [pscustomobject]@{
|
|
name = $taskName
|
|
exists = $false
|
|
state = "missing"
|
|
lastTaskResult = $null
|
|
lastRunTime = $null
|
|
nextRunTime = $null
|
|
}
|
|
}
|
|
}
|
|
|
|
$queueDir = Join-Path $Config.agentRoot "queue"
|
|
if (-not (Test-Path $queueDir)) { New-Item -ItemType Directory -Force -Path $queueDir | Out-Null }
|
|
$pending = @(Get-ChildItem -Path $queueDir -Filter "*.json" -File -ErrorAction SilentlyContinue | Where-Object { $_.Name -notmatch "^(processed|failed|running)-" })
|
|
$processedDir = Join-Path $queueDir "processed"
|
|
$failedDir = Join-Path $queueDir "failed"
|
|
$processedCount = if (Test-Path $processedDir) { @(Get-ChildItem -Path $processedDir -Filter "*.json" -File -ErrorAction SilentlyContinue).Count } else { 0 }
|
|
$failedCount = if (Test-Path $failedDir) { @(Get-ChildItem -Path $failedDir -Filter "*.json" -File -ErrorAction SilentlyContinue).Count } else { 0 }
|
|
$alertIncomingDir = Join-Path $Config.agentRoot "alerts\incoming"
|
|
$alertProcessedDir = Join-Path $Config.agentRoot "alerts\processed"
|
|
$alertFailedDir = Join-Path $Config.agentRoot "alerts\failed"
|
|
$alertPendingCount = if (Test-Path $alertIncomingDir) { @(Get-ChildItem -Path $alertIncomingDir -Filter "*.json" -File -ErrorAction SilentlyContinue).Count } else { 0 }
|
|
$alertProcessedCount = if (Test-Path $alertProcessedDir) { @(Get-ChildItem -Path $alertProcessedDir -Filter "*.json" -File -ErrorAction SilentlyContinue).Count } else { 0 }
|
|
$alertFailedCount = if (Test-Path $alertFailedDir) { @(Get-ChildItem -Path $alertFailedDir -Filter "*.json" -File -ErrorAction SilentlyContinue).Count } else { 0 }
|
|
$problemStats = Get-AgentProblemStats
|
|
|
|
[pscustomobject]@{
|
|
timestamp = (Get-Date -Format o)
|
|
computer = $env:COMPUTERNAME
|
|
user = $env:USERNAME
|
|
agentRoot = $Config.agentRoot
|
|
evidenceDir = $EvidenceDir
|
|
queue = [pscustomobject]@{
|
|
path = $queueDir
|
|
pending = $pending.Count
|
|
processed = $processedCount
|
|
failed = $failedCount
|
|
}
|
|
alerts = [pscustomobject]@{
|
|
incomingPath = $alertIncomingDir
|
|
pending = $alertPendingCount
|
|
processed = $alertProcessedCount
|
|
failed = $alertFailedCount
|
|
}
|
|
problemCounts = [pscustomobject]@{
|
|
path = $problemStats.path
|
|
totalEvents = $problemStats.totalEvents
|
|
totalResolved = $problemStats.totalResolved
|
|
totalFailed = $problemStats.totalFailed
|
|
top = @($problemStats.problems | Select-Object -First 10)
|
|
}
|
|
tasks = $tasks
|
|
evidence = @(
|
|
Get-AgentLatestEvidenceSummary "status" "agent99-Status-*.json"
|
|
Get-AgentLatestEvidenceSummary "performance" "agent99-Perf-*.json"
|
|
Get-AgentLatestEvidenceSummary "self_health" "agent99-SelfCheck-*.json"
|
|
Get-AgentLatestEvidenceSummary "backup" "agent99-BackupCheck-*.json"
|
|
Get-AgentLatestEvidenceSummary "startup_recovery" "agent99-Recover-*.json"
|
|
)
|
|
public = if ($status.exists -and -not $status.parseError -and $status.data) { @($status.data.public | Select-Object url,status,non502,ok) } else { @() }
|
|
hosts = if ($status.exists -and -not $status.parseError -and $status.data) { @($status.data.hosts | Select-Object host,ping) } else { @() }
|
|
performance = if ($perf.exists -and -not $perf.parseError -and $perf.data) { @($perf.data.performance | Select-Object host,severity,loadPerCore,memAvailablePercent,diskUsedPercent,route) } else { @() }
|
|
selfHealth = if ($self.exists -and -not $self.parseError -and $self.data) { $self.data.selfHealth.summary } else { $null }
|
|
backupHealth = if ($backup.exists -and -not $backup.parseError -and $backup.data) { $backup.data.backupHealth.summary } else { $null }
|
|
latestSelfHealthSeverity = if ($self.exists -and -not $self.parseError -and $self.data) { $self.data.selfHealth.severity } else { "unknown" }
|
|
latestBackupSeverity = if ($backup.exists -and -not $backup.parseError -and $backup.data) { $backup.data.backupHealth.severity } else { "unknown" }
|
|
}
|
|
}
|
|
|
|
function Write-AgentDashboard {
|
|
param([object]$Summary)
|
|
|
|
$dashboardDir = Join-Path $Config.agentRoot "dashboard"
|
|
New-Item -ItemType Directory -Force -Path $dashboardDir | Out-Null
|
|
$jsonPath = Join-Path $dashboardDir "agent99-control-center.latest.json"
|
|
$htmlPath = Join-Path $dashboardDir "index.html"
|
|
$Summary | ConvertTo-Json -Depth 12 | Set-Content -Path $jsonPath -Encoding UTF8
|
|
|
|
$taskRows = foreach ($task in @($Summary.tasks)) {
|
|
"<tr><td>$(Convert-AgentHtml $task.name)</td><td>$(Convert-AgentHtml $task.state)</td><td>$(Convert-AgentHtml $task.lastTaskResult)</td><td>$(Convert-AgentHtml $task.lastRunTime)</td><td>$(Convert-AgentHtml $task.nextRunTime)</td></tr>"
|
|
}
|
|
$hostRows = foreach ($hostItem in @($Summary.hosts)) {
|
|
"<tr><td>$(Convert-AgentHtml $hostItem.host)</td><td>$(Convert-AgentHtml $hostItem.ping)</td></tr>"
|
|
}
|
|
$publicRows = foreach ($route in @($Summary.public)) {
|
|
"<tr><td>$(Convert-AgentHtml $route.url)</td><td>$(Convert-AgentHtml $route.status)</td><td>$(Convert-AgentHtml $route.non502)</td><td>$(Convert-AgentHtml $route.ok)</td></tr>"
|
|
}
|
|
$perfRows = foreach ($perfItem in @($Summary.performance)) {
|
|
"<tr><td>$(Convert-AgentHtml $perfItem.host)</td><td>$(Convert-AgentHtml $perfItem.severity)</td><td>$(Convert-AgentHtml $perfItem.loadPerCore)</td><td>$(Convert-AgentHtml $perfItem.memAvailablePercent)</td><td>$(Convert-AgentHtml $perfItem.diskUsedPercent)</td><td>$(Convert-AgentHtml $perfItem.route)</td></tr>"
|
|
}
|
|
$evidenceRows = foreach ($ev in @($Summary.evidence)) {
|
|
"<tr><td>$(Convert-AgentHtml $ev.name)</td><td>$(Convert-AgentHtml $ev.file)</td><td>$(Convert-AgentHtml $ev.ageMinutes)</td><td>$(Convert-AgentHtml $ev.parseError)</td></tr>"
|
|
}
|
|
$problemRows = foreach ($problem in @($Summary.problemCounts.top)) {
|
|
"<tr><td>$(Convert-AgentHtml $problem.key)</td><td>$(Convert-AgentHtml $problem.occurrences)</td><td>$(Convert-AgentHtml $problem.resolved)</td><td>$(Convert-AgentHtml $problem.failed)</td><td>$(Convert-AgentHtml $problem.lastStatus)</td><td>$(Convert-AgentHtml $problem.lastMode)</td></tr>"
|
|
}
|
|
|
|
$html = @"
|
|
<!doctype html>
|
|
<html>
|
|
<head>
|
|
<meta charset="utf-8">
|
|
<meta http-equiv="refresh" content="30">
|
|
<title>Agent99 Control Center</title>
|
|
<style>
|
|
body { font-family: Segoe UI, Arial, sans-serif; margin: 24px; background: #f5f7fb; color: #16202a; }
|
|
h1 { margin: 0 0 6px; font-size: 28px; }
|
|
h2 { margin-top: 26px; font-size: 18px; }
|
|
.meta { color: #526170; margin-bottom: 18px; }
|
|
.grid { display: grid; grid-template-columns: repeat(4, minmax(160px, 1fr)); gap: 12px; }
|
|
.card { background: #fff; border: 1px solid #d8e0ea; border-radius: 8px; padding: 14px; }
|
|
.label { color: #637282; font-size: 12px; text-transform: uppercase; }
|
|
.value { font-size: 22px; margin-top: 4px; font-weight: 650; }
|
|
table { width: 100%; border-collapse: collapse; background: #fff; border: 1px solid #d8e0ea; }
|
|
th, td { padding: 8px 10px; border-bottom: 1px solid #e7edf5; text-align: left; font-size: 13px; }
|
|
th { background: #eef3f8; }
|
|
.ok { color: #087443; }
|
|
.path { font-family: Consolas, monospace; font-size: 12px; color: #465565; }
|
|
</style>
|
|
</head>
|
|
<body>
|
|
<h1>Agent99 Control Center</h1>
|
|
<div class="meta">Last update: $(Convert-AgentHtml $Summary.timestamp) on $(Convert-AgentHtml $Summary.computer) as $(Convert-AgentHtml $Summary.user). Auto-refreshes every 30 seconds.</div>
|
|
<div class="grid">
|
|
<div class="card"><div class="label">Self Health</div><div class="value">$(Convert-AgentHtml $Summary.latestSelfHealthSeverity)</div></div>
|
|
<div class="card"><div class="label">Backup Health</div><div class="value">$(Convert-AgentHtml $Summary.latestBackupSeverity)</div></div>
|
|
<div class="card"><div class="label">Queue Pending</div><div class="value">$(Convert-AgentHtml $Summary.queue.pending)</div></div>
|
|
<div class="card"><div class="label">Alerts Pending</div><div class="value">$(Convert-AgentHtml $Summary.alerts.pending)</div></div>
|
|
<div class="card"><div class="label">Problem Events</div><div class="value">$(Convert-AgentHtml $Summary.problemCounts.totalEvents)</div></div>
|
|
<div class="card"><div class="label">Resolved</div><div class="value">$(Convert-AgentHtml $Summary.problemCounts.totalResolved)</div></div>
|
|
<div class="card"><div class="label">Failed</div><div class="value">$(Convert-AgentHtml $Summary.problemCounts.totalFailed)</div></div>
|
|
</div>
|
|
<p class="path">Agent root: $(Convert-AgentHtml $Summary.agentRoot)<br>Queue: $(Convert-AgentHtml $Summary.queue.path)<br>Alerts: $(Convert-AgentHtml $Summary.alerts.incomingPath)<br>Problem counts: $(Convert-AgentHtml $Summary.problemCounts.path)<br>Evidence: $(Convert-AgentHtml $Summary.evidenceDir)</p>
|
|
<h2>Problem Counts</h2>
|
|
<table><tr><th>Problem</th><th>Occurrences</th><th>Resolved</th><th>Failed</th><th>Last Status</th><th>Last Mode</th></tr>$($problemRows -join "`n")</table>
|
|
<h2>Scheduled Tasks</h2>
|
|
<table><tr><th>Task</th><th>State</th><th>Last Result</th><th>Last Run</th><th>Next Run</th></tr>$($taskRows -join "`n")</table>
|
|
<h2>Hosts</h2>
|
|
<table><tr><th>Host</th><th>Ping</th></tr>$($hostRows -join "`n")</table>
|
|
<h2>Public Routes</h2>
|
|
<table><tr><th>URL</th><th>Status</th><th>Non-502</th><th>OK</th></tr>$($publicRows -join "`n")</table>
|
|
<h2>Performance</h2>
|
|
<table><tr><th>Host</th><th>Severity</th><th>Load/Core</th><th>Mem Avail %</th><th>Disk Used %</th><th>Route</th></tr>$($perfRows -join "`n")</table>
|
|
<h2>Evidence</h2>
|
|
<table><tr><th>Name</th><th>File</th><th>Age Minutes</th><th>Parse Error</th></tr>$($evidenceRows -join "`n")</table>
|
|
</body>
|
|
</html>
|
|
"@
|
|
Set-Content -Path $htmlPath -Value $html -Encoding UTF8
|
|
[pscustomobject]@{
|
|
dashboardDir = $dashboardDir
|
|
htmlPath = $htmlPath
|
|
jsonPath = $jsonPath
|
|
}
|
|
}
|
|
|
|
function Ensure-AgentDesktopShortcuts {
|
|
param([string]$DashboardPath)
|
|
|
|
$created = @()
|
|
$targets = @(
|
|
[pscustomobject]@{ name = "Agent99 Control Center.lnk"; target = $DashboardPath },
|
|
[pscustomobject]@{ name = "Agent99 Submit Request.lnk"; target = (Join-Path $Config.agentRoot "agent99-submit-request.cmd") },
|
|
[pscustomobject]@{ name = "Agent99 Queue.lnk"; target = (Join-Path $Config.agentRoot "queue") },
|
|
[pscustomobject]@{ name = "Agent99 Evidence.lnk"; target = $EvidenceDir }
|
|
)
|
|
$desktopDirs = @(
|
|
[Environment]::GetFolderPath("CommonDesktopDirectory"),
|
|
[Environment]::GetFolderPath("Desktop")
|
|
) | Where-Object { $_ } | Select-Object -Unique
|
|
|
|
foreach ($desktop in $desktopDirs) {
|
|
if (-not (Test-Path $desktop)) { continue }
|
|
foreach ($item in $targets) {
|
|
$shortcutPath = Join-Path $desktop $item.name
|
|
try {
|
|
$shell = New-Object -ComObject WScript.Shell
|
|
$shortcut = $shell.CreateShortcut($shortcutPath)
|
|
$shortcut.TargetPath = $item.target
|
|
$shortcut.WorkingDirectory = Split-Path -Parent $item.target
|
|
$shortcut.Description = "Agent99 local control and evidence shortcut"
|
|
$shortcut.Save()
|
|
$created += $shortcutPath
|
|
} catch {
|
|
Write-AgentLog "desktop_shortcut_failed path=$shortcutPath error=$($_.Exception.Message)"
|
|
}
|
|
}
|
|
}
|
|
$created
|
|
}
|
|
|
|
function Invoke-AgentQueuedCommands {
|
|
$queueDir = Join-Path $Config.agentRoot "queue"
|
|
$processedDir = Join-Path $queueDir "processed"
|
|
$failedDir = Join-Path $queueDir "failed"
|
|
New-Item -ItemType Directory -Force -Path $queueDir, $processedDir, $failedDir | Out-Null
|
|
|
|
$allowedModes = @("Status", "Recover", "StartVMs", "PublicSmoke", "HarborRepair", "AwoooRepair", "Perf", "LoadShed", "SelfCheck", "BackupCheck", "ProviderFreshness")
|
|
$commands = @(Get-ChildItem -Path $queueDir -Filter "*.json" -File -ErrorAction SilentlyContinue |
|
|
Where-Object { $_.Name -notmatch "^(processed|failed|running)-" } |
|
|
Sort-Object LastWriteTime |
|
|
Select-Object -First 3)
|
|
$results = @()
|
|
foreach ($file in $commands) {
|
|
$command = $null
|
|
try {
|
|
$command = Get-Content $file.FullName -Raw | ConvertFrom-Json
|
|
} catch {
|
|
$target = Join-Path $failedDir ("failed-" + $file.Name)
|
|
Move-Item -Force $file.FullName $target
|
|
$results += [pscustomobject]@{ id = $file.BaseName; ok = $false; reason = "invalid_json"; file = $target }
|
|
continue
|
|
}
|
|
|
|
$id = if ($command.id) { [string]$command.id } else { $file.BaseName }
|
|
$modeName = if ($command.mode) { [string]$command.mode } else { "" }
|
|
$controlled = if ($command.PSObject.Properties["controlledApply"]) { [bool]$command.controlledApply } else { $false }
|
|
$commandSource = if ($command.PSObject.Properties["source"]) { [string]$command.source } else { "queue" }
|
|
$commandReason = if ($command.PSObject.Properties["reason"]) { [string]$command.reason } else { "operator_instruction" }
|
|
$commandInstruction = if ($command.PSObject.Properties["instruction"]) { [string]$command.instruction } else { "" }
|
|
$alertId = if ($command.PSObject.Properties["alertId"]) { [string]$command.alertId } else { $null }
|
|
$alertSource = if ($command.PSObject.Properties["alertSource"]) { [string]$command.alertSource } else { $null }
|
|
$alertKind = if ($command.PSObject.Properties["alertKind"]) { [string]$command.alertKind } else { $null }
|
|
$alertSeverity = if ($command.PSObject.Properties["alertSeverity"]) { [string]$command.alertSeverity } else { $null }
|
|
$alertService = if ($command.PSObject.Properties["alertService"]) { [string]$command.alertService } else { $null }
|
|
$alertHost = if ($command.PSObject.Properties["alertHost"]) { [string]$command.alertHost } else { $null }
|
|
$replyChatId = if ($command.PSObject.Properties["replyChatId"]) { [string]$command.replyChatId } else { "" }
|
|
if ($modeName -notin $allowedModes) {
|
|
$target = Join-Path $failedDir ("failed-" + $file.Name)
|
|
Move-Item -Force $file.FullName $target
|
|
$result = [pscustomobject]@{ id = $id; mode = $modeName; source = $commandSource; instruction = $commandInstruction; ok = $false; reason = "mode_not_allowlisted"; file = $target }
|
|
Record-AgentEvent "operator_command_rejected" "warning" "id=$id mode=$modeName reason=mode_not_allowlisted" $result -Alert
|
|
$results += $result
|
|
continue
|
|
}
|
|
|
|
$runningPath = Join-Path $queueDir ("running-" + $file.Name)
|
|
Move-Item -Force $file.FullName $runningPath
|
|
$startTime = Get-Date
|
|
$launcher = Join-Path $Config.agentRoot "agent99-run.ps1"
|
|
$stdoutPath = Join-Path $EvidenceDir "agent99-command-$id.out"
|
|
$stderrPath = Join-Path $EvidenceDir "agent99-command-$id.err"
|
|
$args = @("-NoProfile", "-ExecutionPolicy", "Bypass", "-File", $launcher, "-Mode", $modeName)
|
|
if ($controlled) { $args += "-ControlledApply" }
|
|
$process = Start-Process -FilePath "powershell.exe" -ArgumentList $args -NoNewWindow -RedirectStandardOutput $stdoutPath -RedirectStandardError $stderrPath -PassThru
|
|
$completed = $process.WaitForExit(600000)
|
|
if (-not $completed) {
|
|
$process.Kill()
|
|
$process.WaitForExit(2000) | Out-Null
|
|
}
|
|
$process.WaitForExit() | Out-Null
|
|
$process.Refresh()
|
|
$latest = Get-AgentLatestEvidence "agent99-$modeName-*.json"
|
|
$exitCode = if (-not $completed) {
|
|
-2
|
|
} elseif ($null -ne $process.ExitCode) {
|
|
$process.ExitCode
|
|
} elseif ($latest.exists -and -not $latest.parseError) {
|
|
0
|
|
} else {
|
|
-3
|
|
}
|
|
$result = [pscustomobject]@{
|
|
id = $id
|
|
mode = $modeName
|
|
controlledApply = $controlled
|
|
source = $commandSource
|
|
reason = $commandReason
|
|
instruction = $commandInstruction
|
|
alertId = $alertId
|
|
alertSource = $alertSource
|
|
alertKind = $alertKind
|
|
alertSeverity = $alertSeverity
|
|
alertService = $alertService
|
|
alertHost = $alertHost
|
|
ok = [bool]($exitCode -eq 0)
|
|
exitCode = $exitCode
|
|
durationSeconds = [math]::Round(((Get-Date) - $startTime).TotalSeconds, 1)
|
|
evidence = $latest.path
|
|
stdout = $stdoutPath
|
|
stderr = $stderrPath
|
|
}
|
|
$problem = Update-AgentProblemKnowledge $result
|
|
$result | Add-Member -MemberType NoteProperty -Name problem -Value $problem -Force
|
|
$donePath = Join-Path $processedDir ("processed-$id.json")
|
|
$result | ConvertTo-Json -Depth 8 | Set-Content -Path $donePath -Encoding UTF8
|
|
Remove-Item $runningPath -Force -ErrorAction SilentlyContinue | Out-Null
|
|
$recordData = $result | Select-Object *
|
|
if ($replyChatId) {
|
|
$recordData | Add-Member -MemberType NoteProperty -Name replyChatId -Value $replyChatId -Force
|
|
}
|
|
Record-AgentEvent "operator_command_result" $(if ($exitCode -eq 0) { "info" } else { "critical" }) "id=$id mode=$modeName exitCode=$exitCode evidence=$($latest.path)" $recordData -Alert
|
|
$results += $result
|
|
}
|
|
$results
|
|
}
|
|
|
|
function Invoke-AgentControlTick {
|
|
$processed = Invoke-AgentQueuedCommands
|
|
$summary = Get-AgentDashboardSummary
|
|
$dashboard = Write-AgentDashboard $summary
|
|
$shortcuts = Ensure-AgentDesktopShortcuts $dashboard.htmlPath
|
|
$pendingCount = if ($summary.queue) { $summary.queue.pending } else { 0 }
|
|
$result = [pscustomobject]@{
|
|
ok = $true
|
|
dashboardPath = $dashboard.htmlPath
|
|
dashboardJson = $dashboard.jsonPath
|
|
shortcutCount = @($shortcuts).Count
|
|
shortcuts = $shortcuts
|
|
processedCount = @($processed).Count
|
|
processed = $processed
|
|
pendingCount = $pendingCount
|
|
}
|
|
Record-AgentEvent "agent_control_tick" "info" "dashboard=$($dashboard.htmlPath) processed=$(@($processed).Count) pending=$pendingCount" $result -NoAlert
|
|
$result
|
|
}
|
|
|
|
function Invoke-AgentProviderFreshnessDirectReadback {
|
|
param(
|
|
[int]$LookbackMinutes
|
|
)
|
|
|
|
$baseUrl = "https://awoooi.wooo.work/api/v1/platform/events/dossier/recurrence?project_id=awoooi&limit=20"
|
|
$startedAt = Get-Date
|
|
try {
|
|
$response = Invoke-RestMethod -Method Get -Uri $baseUrl -TimeoutSec 12
|
|
$summary = if ($response -and $response.PSObject.Properties["summary"]) { $response.summary } else { $null }
|
|
$latest = if ($summary -and $summary.PSObject.Properties["latest_received_at"]) { $summary.latest_received_at } else { $null }
|
|
$sourceTotal = if ($summary -and $summary.PSObject.Properties["source_event_total"]) { [int]$summary.source_event_total } else { 0 }
|
|
$itemsCount = if ($response -and $response.PSObject.Properties["items"]) { @($response.items).Count } else { 0 }
|
|
return [pscustomobject]@{
|
|
ok = $true
|
|
verifier = "awoooi-platform-events-recurrence"
|
|
directReference = $baseUrl
|
|
providerWindowMinutes = $LookbackMinutes
|
|
latestReceivedAt = $latest
|
|
sourceEventTotal = $sourceTotal
|
|
recurrenceGroupTotal = if ($summary -and $summary.PSObject.Properties["recurrence_group_total"]) { [int]$summary.recurrence_group_total } else { 0 }
|
|
itemsCount = $itemsCount
|
|
httpStatus = 200
|
|
durationSeconds = [math]::Round(((Get-Date) - $startedAt).TotalSeconds, 2)
|
|
checkedAt = (Get-Date -Format o)
|
|
}
|
|
} catch {
|
|
return [pscustomobject]@{
|
|
ok = $false
|
|
verifier = "awoooi-platform-events-recurrence"
|
|
directReference = $baseUrl
|
|
providerWindowMinutes = $LookbackMinutes
|
|
latestReceivedAt = $null
|
|
sourceEventTotal = 0
|
|
recurrenceGroupTotal = 0
|
|
itemsCount = 0
|
|
httpStatus = $null
|
|
durationSeconds = [math]::Round(((Get-Date) - $startedAt).TotalSeconds, 2)
|
|
checkedAt = (Get-Date -Format o)
|
|
error = ($_ | Out-String).Trim()
|
|
}
|
|
}
|
|
}
|
|
|
|
function Invoke-AgentProviderFreshnessTriage {
|
|
$providerConfig = if ($Config.PSObject.Properties["providerFreshness"]) { $Config.providerFreshness } else { $null }
|
|
$lookbackMinutes = if ($providerConfig -and $providerConfig.PSObject.Properties["lookbackMinutes"]) { [int]$providerConfig.lookbackMinutes } else { 120 }
|
|
$since = (Get-Date).AddMinutes(-1 * $lookbackMinutes)
|
|
$candidateId = "provider-freshness-" + (Get-Date -Format "yyyyMMdd-HHmmss")
|
|
$candidateDir = Join-Path $Config.agentRoot "state\provider-freshness-candidates"
|
|
New-Item -ItemType Directory -Force -Path $candidateDir | Out-Null
|
|
$candidatePath = Join-Path $candidateDir "$candidateId.json"
|
|
|
|
$patterns = @(
|
|
"provider_freshness_signal",
|
|
"source_provider_freshness_triage",
|
|
"raw_signal_normalized",
|
|
"controlled_runtime_candidate",
|
|
"freshness",
|
|
"last_seen",
|
|
"last seen",
|
|
"ingestion"
|
|
)
|
|
$scanDirs = @(
|
|
(Join-Path $Config.agentRoot "alerts\incoming"),
|
|
(Join-Path $Config.agentRoot "alerts\processed"),
|
|
(Join-Path $Config.agentRoot "queue"),
|
|
(Join-Path $Config.agentRoot "queue\processed"),
|
|
(Join-Path $Config.agentRoot "requests")
|
|
) | Where-Object { Test-Path $_ } | Select-Object -Unique
|
|
|
|
$hits = @()
|
|
foreach ($dir in $scanDirs) {
|
|
$files = @(Get-ChildItem -LiteralPath $dir -Recurse -File -ErrorAction SilentlyContinue |
|
|
Where-Object { $_.LastWriteTime -ge $since -and $_.Length -lt 2097152 })
|
|
foreach ($file in $files) {
|
|
$matchedPatterns = @()
|
|
foreach ($pattern in $patterns) {
|
|
if (Select-String -LiteralPath $file.FullName -Pattern $pattern -SimpleMatch -Quiet -ErrorAction SilentlyContinue) {
|
|
$matchedPatterns += $pattern
|
|
}
|
|
}
|
|
if (@($matchedPatterns).Count -gt 0) {
|
|
$excerpt = ""
|
|
$firstHit = Select-String -LiteralPath $file.FullName -Pattern $matchedPatterns[0] -SimpleMatch -ErrorAction SilentlyContinue | Select-Object -First 1
|
|
if ($firstHit) {
|
|
$excerpt = [string]$firstHit.Line
|
|
$excerpt = $excerpt -replace '("?(telegram|token|chat|authorization|password)[^,}\r\n]*"?\s*:\s*)"[^"]+"', '$1"[redacted]"'
|
|
if ($excerpt.Length -gt 500) { $excerpt = $excerpt.Substring(0, 500) + "...(truncated)" }
|
|
}
|
|
$hits += [pscustomobject]@{
|
|
path = $file.FullName
|
|
lastWrite = $file.LastWriteTime.ToString("o")
|
|
patterns = @($matchedPatterns)
|
|
excerpt = $excerpt
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
$targets = @()
|
|
if ($providerConfig -and $providerConfig.targets) {
|
|
foreach ($target in @($providerConfig.targets)) {
|
|
$targets += [pscustomobject]@{
|
|
name = if ($target.name) { [string]$target.name } else { "provider_freshness_signal" }
|
|
providerWindowMinutes = if ($target.PSObject.Properties["providerWindowMinutes"]) { $target.providerWindowMinutes } else { $null }
|
|
directReference = if ($target.directReference) { [string]$target.directReference } else { $null }
|
|
verifier = if ($target.verifier) { [string]$target.verifier } else { "agent99-provider-freshness-triage" }
|
|
}
|
|
}
|
|
}
|
|
if (@($targets).Count -eq 0) {
|
|
$targets += [pscustomobject]@{
|
|
name = "provider_freshness_signal"
|
|
providerWindowMinutes = $null
|
|
directReference = $null
|
|
verifier = "agent99-provider-freshness-triage"
|
|
}
|
|
}
|
|
|
|
$verifierReadback = Invoke-AgentProviderFreshnessDirectReadback -LookbackMinutes $lookbackMinutes
|
|
$directReference = if ($verifierReadback -and $verifierReadback.PSObject.Properties["directReference"]) { $verifierReadback.directReference } else { $null }
|
|
$latestSeen = if ($verifierReadback -and $verifierReadback.PSObject.Properties["latestReceivedAt"]) { $verifierReadback.latestReceivedAt } else { $null }
|
|
$missingFields = @()
|
|
if (-not $lookbackMinutes) { $missingFields += "providerWindow" }
|
|
if (-not $latestSeen) { $missingFields += "lastSeen" }
|
|
if (-not $directReference) { $missingFields += "directReference" }
|
|
if (-not ($verifierReadback -and $verifierReadback.ok -eq $true)) { $missingFields += "verifierReadback" }
|
|
$status = if ($verifierReadback -and $verifierReadback.ok -eq $true -and $latestSeen) {
|
|
"direct_readback_ok"
|
|
} elseif ($verifierReadback -and $verifierReadback.ok -eq $true) {
|
|
"direct_readback_ok_no_source_events"
|
|
} elseif (@($hits).Count -gt 0) {
|
|
"candidate_created_waiting_direct_readback"
|
|
} else {
|
|
"candidate_created_no_alert_body_seen"
|
|
}
|
|
$candidate = [pscustomobject]@{
|
|
timestamp = (Get-Date -Format o)
|
|
candidateId = $candidateId
|
|
target = "provider_freshness_signal"
|
|
status = $status
|
|
lookbackMinutes = $lookbackMinutes
|
|
since = $since.ToString("o")
|
|
hitCount = @($hits).Count
|
|
hits = @($hits | Sort-Object lastWrite -Descending | Select-Object -First 20)
|
|
targets = $targets
|
|
missingFields = $missingFields
|
|
providerWindowMinutes = $lookbackMinutes
|
|
lastSeen = $latestSeen
|
|
directReference = $directReference
|
|
verifierReadback = $verifierReadback
|
|
requiredVerifierReadback = @("provider_window", "last_seen", "direct_reference", "source_trace_or_sentry_signoz_ref")
|
|
noFalseGreenRisk = $true
|
|
allowedActions = @("create_candidate", "read_existing_evidence", "provider_freshness_direct_readback", "request_direct_reference", "write_km")
|
|
prohibitedActions = @("provider_switch", "paid_model_call", "env_change", "reload")
|
|
candidatePath = $candidatePath
|
|
}
|
|
$candidate | ConvertTo-Json -Depth 10 | Set-Content -Path $candidatePath -Encoding UTF8
|
|
Write-AgentLog "provider_freshness candidate=$candidateId hits=$(@($hits).Count) status=$status missing=$($missingFields -join ',') path=$candidatePath"
|
|
$eventSeverity = if (@($missingFields).Count -gt 0) { "warning" } else { "info" }
|
|
Record-AgentEvent "provider_freshness_triage" $eventSeverity "target=provider_freshness_signal status=$status missing=$($missingFields -join ',')" $candidate -Alert
|
|
$candidate
|
|
}
|
|
|
|
function Test-HostReachability {
|
|
$results = @()
|
|
foreach ($hostIp in $Config.hosts) {
|
|
$ping = Test-Connection -ComputerName $hostIp -Count 1 -Quiet -ErrorAction SilentlyContinue
|
|
Write-AgentLog "host host=$hostIp ping=$ping"
|
|
$results += [pscustomobject]@{
|
|
host = $hostIp
|
|
ping = [bool]$ping
|
|
}
|
|
}
|
|
$results
|
|
}
|
|
|
|
function Test-PublicRoutes {
|
|
$results = @()
|
|
foreach ($url in $Config.publicUrls) {
|
|
try {
|
|
$output = & curl.exe --location --head --max-time 12 $url 2>&1
|
|
$text = $output -join "`n"
|
|
$matches = [regex]::Matches($text, "HTTP/\S+\s+(\d+)")
|
|
$status = if ($matches.Count -gt 0) { [int]$matches[$matches.Count - 1].Groups[1].Value } else { $null }
|
|
$non502 = ($null -ne $status -and $status -ne 502)
|
|
$ok = ($null -ne $status -and $status -ge 200 -and $status -lt 500)
|
|
Write-AgentLog "public url=$url status=$status non502=$non502 ok=$ok"
|
|
$results += [pscustomobject]@{
|
|
url = $url
|
|
status = $status
|
|
non502 = $non502
|
|
ok = $ok
|
|
}
|
|
} catch {
|
|
Write-AgentLog "public url=$url error=$($_.Exception.Message)"
|
|
$results += [pscustomobject]@{
|
|
url = $url
|
|
status = $null
|
|
non502 = $false
|
|
ok = $false
|
|
}
|
|
}
|
|
}
|
|
$results
|
|
}
|
|
|
|
function Test-AiServices {
|
|
$results = @()
|
|
foreach ($svc in @($Config.aiServices)) {
|
|
if (-not $svc.url) { continue }
|
|
$name = if ($svc.name) { [string]$svc.name } else { [string]$svc.url }
|
|
try {
|
|
$output = & curl.exe --max-time 12 -sS $svc.url 2>&1
|
|
$text = ($output -join "`n").Trim()
|
|
$statusValue = $null
|
|
try {
|
|
$json = $text | ConvertFrom-Json
|
|
if ($json.status) { $statusValue = [string]$json.status }
|
|
} catch {
|
|
$statusValue = $null
|
|
}
|
|
$expectedStatus = if ($svc.expectedStatus) { [string]$svc.expectedStatus } else { "healthy" }
|
|
$ok = ($statusValue -eq $expectedStatus)
|
|
$severity = if ($ok) { "ok" } elseif ($svc.critical) { "critical" } else { "warning" }
|
|
$result = [pscustomobject]@{
|
|
name = $name
|
|
url = $svc.url
|
|
host = $svc.host
|
|
ok = $ok
|
|
severity = $severity
|
|
status = $statusValue
|
|
expectedStatus = $expectedStatus
|
|
output = $text
|
|
}
|
|
Write-AgentLog "ai_service name=$name ok=$ok severity=$severity status=$statusValue url=$($svc.url)"
|
|
if (-not $ok) {
|
|
Record-AgentEvent "ai_service_unhealthy" $severity "name=$name status=$statusValue expected=$expectedStatus url=$($svc.url)" $result -Alert
|
|
}
|
|
$results += $result
|
|
} catch {
|
|
$severity = if ($svc.critical) { "critical" } else { "warning" }
|
|
$result = [pscustomobject]@{
|
|
name = $name
|
|
url = $svc.url
|
|
host = $svc.host
|
|
ok = $false
|
|
severity = $severity
|
|
status = $null
|
|
expectedStatus = if ($svc.expectedStatus) { [string]$svc.expectedStatus } else { "healthy" }
|
|
output = $_.Exception.Message
|
|
}
|
|
Write-AgentLog "ai_service name=$name error=$($_.Exception.Message)"
|
|
Record-AgentEvent "ai_service_unhealthy" $severity "name=$name error=$($_.Exception.Message) url=$($svc.url)" $result -Alert
|
|
$results += $result
|
|
}
|
|
}
|
|
$results
|
|
}
|
|
|
|
function Get-HarborState {
|
|
$inspectCommand = "docker inspect --format '{{.Name}}|{{.State.Status}}|{{if .State.Health}}{{.State.Health.Status}}{{else}}none{{end}}' harbor-core redis registry"
|
|
$inspect = Invoke-SshText "192.168.0.110" $inspectCommand 30 1
|
|
$containerStates = @{}
|
|
foreach ($line in @(([string]$inspect.output) -split "`n")) {
|
|
$trimmed = $line.Trim()
|
|
if (-not $trimmed) { continue }
|
|
$parts = @($trimmed -split "\|")
|
|
if ($parts.Count -lt 3) { continue }
|
|
$name = $parts[0].Trim().TrimStart("/")
|
|
$containerStates[$name] = [pscustomobject]@{
|
|
name = $name
|
|
status = $parts[1].Trim()
|
|
health = $parts[2].Trim()
|
|
}
|
|
}
|
|
$core = $containerStates["harbor-core"]
|
|
$redis = $containerStates["redis"]
|
|
$registry = $containerStates["registry"]
|
|
$coreLine = if ($core) { "$($core.name) $($core.status) $($core.health)" } else { "harbor-core missing" }
|
|
$redisLine = if ($redis) { "$($redis.name) $($redis.status) $($redis.health)" } else { "redis missing" }
|
|
$registryLine = if ($registry) { "$($registry.name) $($registry.status) $($registry.health)" } else { "registry missing" }
|
|
|
|
$state = [pscustomobject]@{
|
|
coreHealthy = ($core -and $core.status -eq "running" -and $core.health -eq "healthy")
|
|
redisUp = ($redis -and $redis.status -eq "running")
|
|
registryHealthy = ($registry -and $registry.status -eq "running" -and $registry.health -eq "healthy")
|
|
coreOutput = $coreLine
|
|
redisOutput = $redisLine
|
|
registryOutput = $registryLine
|
|
inspectExitCode = $inspect.exitCode
|
|
healthOutput = $inspect.output
|
|
inspectOutput = "[redacted: raw docker inspect omitted]"
|
|
}
|
|
Write-AgentLog "harbor coreHealthy=$($state.coreHealthy) redisUp=$($state.redisUp) registryHealthy=$($state.registryHealthy)"
|
|
$state
|
|
}
|
|
|
|
function Get-AwoooState {
|
|
$nestedSsh = "ssh -o BatchMode=yes -o StrictHostKeyChecking=accept-new -o NumberOfPasswordPrompts=0 -o ConnectTimeout=8 -l $SshUser 192.168.0.120"
|
|
$readback = Invoke-SshText "192.168.0.110" "$nestedSsh sudo -n kubectl get deploy,pods -n awoooi-prod" 60 1
|
|
|
|
$state = [pscustomobject]@{
|
|
deployReady = (
|
|
$readback.output -match "deployment.apps/awoooi-api\s+1/1" -and
|
|
$readback.output -match "deployment.apps/awoooi-web\s+2/2" -and
|
|
$readback.output -match "deployment.apps/awoooi-worker\s+1/1"
|
|
)
|
|
imagePullFailure = ($readback.output -match "ImagePullBackOff|ErrImagePull")
|
|
crashFailure = ($readback.output -match "CrashLoopBackOff|Pending")
|
|
deployOutput = $readback.output
|
|
podOutput = $readback.output
|
|
readbackExitCode = $readback.exitCode
|
|
}
|
|
Write-AgentLog "awoooi deployReady=$($state.deployReady) imagePullFailure=$($state.imagePullFailure) crashFailure=$($state.crashFailure)"
|
|
$state
|
|
}
|
|
|
|
function Find-Vmrun {
|
|
if ($Config.vmrunPath -and (Test-Path $Config.vmrunPath)) {
|
|
return $Config.vmrunPath
|
|
}
|
|
|
|
foreach ($candidate in @(
|
|
"C:\Program Files (x86)\VMware\VMware Workstation\vmrun.exe",
|
|
"C:\Program Files\VMware\VMware Workstation\vmrun.exe"
|
|
)) {
|
|
if (Test-Path $candidate) { return $candidate }
|
|
}
|
|
|
|
return $null
|
|
}
|
|
|
|
function Start-ConfiguredVMs {
|
|
$vmrun = Find-Vmrun
|
|
$results = @()
|
|
|
|
if (-not $vmrun) {
|
|
Write-AgentLog "vmrun missing"
|
|
return @([pscustomobject]@{ ok = $false; reason = "vmrun_missing" })
|
|
}
|
|
|
|
foreach ($vm in $Config.vms) {
|
|
if (-not $vm.vmx) {
|
|
$results += [pscustomobject]@{ name = $vm.name; ok = $false; skipped = $true; reason = "missing_vmx_path" }
|
|
continue
|
|
}
|
|
|
|
$hostPing = $false
|
|
if ($vm.host) {
|
|
$hostPing = Test-Connection -ComputerName $vm.host -Count 1 -Quiet -ErrorAction SilentlyContinue
|
|
}
|
|
|
|
if ($hostPing) {
|
|
$results += [pscustomobject]@{ name = $vm.name; ok = $true; skipped = $true; reason = "host_already_reachable" }
|
|
continue
|
|
}
|
|
|
|
if (-not $ControlledApply) {
|
|
$results += [pscustomobject]@{ name = $vm.name; ok = $false; skipped = $true; reason = "controlled_apply_required" }
|
|
continue
|
|
}
|
|
|
|
Write-AgentLog "CONTROLLED_APPLY vm_start name=$($vm.name) vmx=$($vm.vmx)"
|
|
$output = & $vmrun start $vm.vmx nogui 2>&1
|
|
$results += [pscustomobject]@{
|
|
name = $vm.name
|
|
ok = ($LASTEXITCODE -eq 0)
|
|
skipped = $false
|
|
output = ($output -join "`n")
|
|
}
|
|
}
|
|
|
|
$results
|
|
}
|
|
|
|
function Repair-Harbor {
|
|
param([object]$Harbor)
|
|
if (-not $ControlledApply) {
|
|
Write-AgentLog "harbor repair skipped controlled_apply=false"
|
|
return $false
|
|
}
|
|
|
|
$changed = $false
|
|
if (-not $Harbor.redisUp) {
|
|
Write-AgentLog "CONTROLLED_APPLY harbor restart redis"
|
|
Invoke-SshText "192.168.0.110" "docker restart redis" | Out-Null
|
|
$changed = $true
|
|
}
|
|
|
|
if (-not $Harbor.coreHealthy) {
|
|
Write-AgentLog "CONTROLLED_APPLY harbor restart harbor-core"
|
|
Invoke-SshText "192.168.0.110" "docker restart harbor-core" | Out-Null
|
|
$changed = $true
|
|
}
|
|
|
|
return $changed
|
|
}
|
|
|
|
function Repair-Awooo {
|
|
param([object]$Awooo, [object]$Harbor)
|
|
if (-not $ControlledApply) {
|
|
Write-AgentLog "awoooi repair skipped controlled_apply=false"
|
|
return $false
|
|
}
|
|
if (-not $Harbor.coreHealthy) {
|
|
Write-AgentLog "awoooi repair skipped harbor_core_not_healthy"
|
|
return $false
|
|
}
|
|
if (-not $Awooo.imagePullFailure) {
|
|
Write-AgentLog "awoooi repair skipped no_image_pull_failure"
|
|
return $false
|
|
}
|
|
|
|
Write-AgentLog "CONTROLLED_APPLY awoooi delete failed pods"
|
|
$nestedSsh = "ssh -o BatchMode=yes -o StrictHostKeyChecking=accept-new -o NumberOfPasswordPrompts=0 -o ConnectTimeout=8 -l $SshUser 192.168.0.120"
|
|
Invoke-SshText "192.168.0.110" "$nestedSsh sudo -n kubectl delete pod -n awoooi-prod --selector app=awoooi-api" 25 | Out-Null
|
|
Invoke-SshText "192.168.0.110" "$nestedSsh sudo -n kubectl delete pod -n awoooi-prod --selector app=awoooi-web" 25 | Out-Null
|
|
Invoke-SshText "192.168.0.110" "$nestedSsh sudo -n kubectl delete pod -n awoooi-prod --selector app=awoooi-worker" 25 | Out-Null
|
|
return $true
|
|
}
|
|
|
|
Write-AgentLog "agent99_start mode=$Mode controlledApply=$ControlledApply config=$ConfigPath"
|
|
Record-AgentEvent "agent_start" "info" "mode=$Mode controlledApply=$ControlledApply host=$env:COMPUTERNAME" $null -Alert
|
|
|
|
$vmResults = @()
|
|
$hostResults = @()
|
|
$harbor = $null
|
|
$awooo = $null
|
|
$public = @()
|
|
$aiServices = @()
|
|
$performance = @()
|
|
$loadShedding = @()
|
|
$selfHealth = $null
|
|
$backupHealth = $null
|
|
$providerFreshness = $null
|
|
$controlTick = $null
|
|
|
|
if ($Mode -in @("Recover", "StartVMs")) {
|
|
$vmResults = Start-ConfiguredVMs
|
|
}
|
|
|
|
if ($Mode -in @("Status", "Recover")) {
|
|
$hostResults = Test-HostReachability
|
|
}
|
|
|
|
if ($Mode -in @("Status", "Recover", "HarborRepair")) {
|
|
$harbor = Get-HarborState
|
|
}
|
|
|
|
if ($Mode -in @("Recover", "HarborRepair") -and $harbor) {
|
|
$harborChanged = Repair-Harbor $harbor
|
|
if ($harborChanged) {
|
|
Start-Sleep -Seconds 3
|
|
$harbor = Get-HarborState
|
|
}
|
|
}
|
|
|
|
if ($Mode -in @("Status", "Recover", "AwoooRepair")) {
|
|
$awooo = Get-AwoooState
|
|
}
|
|
|
|
if ($Mode -in @("Recover", "AwoooRepair") -and $awooo) {
|
|
if (-not $harbor) { $harbor = Get-HarborState }
|
|
$awoooChanged = Repair-Awooo $awooo $harbor
|
|
if ($awoooChanged) {
|
|
Start-Sleep -Seconds 3
|
|
$awooo = Get-AwoooState
|
|
}
|
|
}
|
|
|
|
if ($Mode -in @("Status", "Recover", "PublicSmoke")) {
|
|
$public = Test-PublicRoutes
|
|
}
|
|
|
|
if ($Mode -in @("Status", "Recover", "Perf")) {
|
|
$aiServices = Test-AiServices
|
|
}
|
|
|
|
if ($Mode -in @("Status", "Recover", "Perf", "LoadShed")) {
|
|
$suppressPerfAlerts = [bool]($ControlledApply -and $Mode -in @("Perf", "LoadShed"))
|
|
$performance = Get-HostPerformance -SuppressAlerts:$suppressPerfAlerts
|
|
}
|
|
|
|
if ($Mode -in @("Perf", "LoadShed")) {
|
|
$loadShedding = Invoke-LoadShedding $performance
|
|
if ($ControlledApply) {
|
|
Record-PerformanceIssuesWithoutRemediation $performance $loadShedding
|
|
}
|
|
}
|
|
|
|
if ($Mode -in @("SelfCheck")) {
|
|
$selfHealth = Test-AgentSelfHealth
|
|
}
|
|
|
|
if ($Mode -in @("BackupCheck")) {
|
|
$backupHealth = Test-AgentBackupHealth
|
|
}
|
|
|
|
if ($Mode -in @("ProviderFreshness")) {
|
|
$providerFreshness = Invoke-AgentProviderFreshnessTriage
|
|
}
|
|
|
|
if ($Mode -in @("ControlTick")) {
|
|
$controlTick = Invoke-AgentControlTick
|
|
}
|
|
|
|
$publicFailures = @($public | Where-Object { $_ -and -not $_.ok }).Count
|
|
$aiFailures = @($aiServices | Where-Object { $_ -and -not $_.ok }).Count
|
|
$perfCritical = @($performance | Where-Object { $_ -and $_.severity -eq "critical" }).Count
|
|
$perfWarning = @($performance | Where-Object { $_ -and $_.severity -eq "warning" }).Count
|
|
$selfHealthCritical = if ($selfHealth -and $selfHealth.severity -eq "critical") { 1 } else { 0 }
|
|
$selfHealthWarning = if ($selfHealth -and $selfHealth.severity -eq "warning") { 1 } else { 0 }
|
|
$backupCritical = if ($backupHealth -and $backupHealth.severity -eq "critical") { 1 } else { 0 }
|
|
$backupWarning = if ($backupHealth -and $backupHealth.severity -eq "warning") { 1 } else { 0 }
|
|
$providerFreshnessWarning = if ($providerFreshness) { 1 } else { 0 }
|
|
$summarySeverity = if ($publicFailures -gt 0 -or $aiFailures -gt 0 -or $perfCritical -gt 0 -or $selfHealthCritical -gt 0 -or $backupCritical -gt 0) { "critical" } elseif ($perfWarning -gt 0 -or $selfHealthWarning -gt 0 -or $backupWarning -gt 0 -or $providerFreshnessWarning -gt 0) { "warning" } else { "info" }
|
|
Record-AgentEvent "agent_complete" $summarySeverity "mode=$Mode publicFailures=$publicFailures aiFailures=$aiFailures perfCritical=$perfCritical perfWarning=$perfWarning selfHealth=$($selfHealth.severity) backupHealth=$($backupHealth.severity) providerFreshness=$($providerFreshness.status) evidence=$jsonPath" $null -NoAlert
|
|
|
|
$result = [pscustomobject]@{
|
|
timestamp = (Get-Date -Format o)
|
|
mode = $Mode
|
|
controlledApply = [bool]$ControlledApply
|
|
vmResults = $vmResults
|
|
hosts = $hostResults
|
|
harbor = $harbor
|
|
awoooi = $awooo
|
|
public = $public
|
|
aiServices = $aiServices
|
|
performance = $performance
|
|
loadShedding = $loadShedding
|
|
selfHealth = $selfHealth
|
|
backupHealth = $backupHealth
|
|
providerFreshness = $providerFreshness
|
|
controlTick = $controlTick
|
|
events = $script:Events
|
|
telegram = $script:TelegramAttempts
|
|
evidenceLog = $logPath
|
|
}
|
|
|
|
$result | ConvertTo-Json -Depth 8 | Set-Content -Encoding UTF8 $jsonPath
|
|
Write-AgentLog "agent99_complete json=$jsonPath"
|
|
|
|
exit 0
|