esn-support-bash-snippets/legacy/exitrace/lib/log_analyzer.awk

#!/usr/bin/awk -f
# log_analyzer.awk - Exim log analyzer for exitrace
# Maintainer: ljp
# Processes Exim logs and classifies delivery status per Message-ID

function trim(s){ sub(/^[ \t\r\n]+/, "", s); sub(/[ \t\r\n]+$/, "", s); return s }
function add_rcpt(id, rc){ if (!(rc in rcpts[id])) rcpts[id][rc]=1 }
function incr(a, k){ a[k]++; return a[k] }

BEGIN { FS=" " }

{
  raw=$0

  # Zeit + ID detektieren
  match(raw, /[0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}:[0-9]{2} [A-Za-z0-9-]+/, t)
  if (!t[0]) next
  split(t[0], p, " ")
  ts = p[1] " " p[2]
  
  # Extrahiere Rest der Zeile nach Timestamp für weitere Verarbeitung
  raw_after = raw
  sub(/^[^:]*:[^:]*:[^ ]+ /, "", raw_after)
  msgid = $1
  if (!msgid) next

  if (MSGIDREQ != "" && msgid != MSGIDREQ) next

  # Init Times
  if (!(msgid in first_ts) || ts < first_ts[msgid]) first_ts[msgid]=ts
  if (!(msgid in last_ts)  || ts > last_ts[msgid])  last_ts[msgid]=ts

  # Sender extrahieren
  if (index($0, "<=") == 1 || $2 == "<=") {
    if (match($0, / from <[^>]+>/, m)) {
      s=m[0]; gsub(/.* from </, "", s); gsub(/>.*/, "", s)
      from=trim(s)
    } else if (match($0, /<= [^ ]+@[^ ]+/, m2)) {
      split(m2[0], a, " "); from=trim(a[2])
    } else { from="" }
    if (from != "") {
      from_addr[msgid]=from
      if (FROMREQ != "" && from != FROMREQ) from_mismatch[msgid]=1
    }
  }

  # Empfänger extrahieren
  rcpt=""
  if (match($0, / for <[^>]+>/, r)) {
    rcpt=r[0]; gsub(/ for </, "", rcpt); gsub(/>/, "", rcpt); rcpt=trim(rcpt)
  } else if (match($0, /\*\* [^ ]+@[^ ]+/, r2)) {
    split(r2[0], b, " "); rcpt=trim(b[2])
  }
  if (rcpt != "") add_rcpt(msgid, rcpt)

  # Klassify line types
  is_delivery = index($0, " => ")>0
  is_fail     = index($0, " ** ")>0
  is_remote   = (index($0, " T=remote_smtp")>0 || index($0, " T=remote_smtp_dkim")>0)

  # Success detection: C="250 ..." oder "250 " in C=...
  is_success = 0
  if (is_delivery && is_remote) {
    if (match($0, /C="[^"]*250[^"]*"/)) is_success=1
  }

  # Deferred detection
  # typische Phrasen: "retry time not reached", "temporarily", "Connection timed out", "have been failing"
  is_defer = 0
  if (is_remote && (index($0, "retry time not reached")>0 || index($0, "temporar")>0 || index($0, "timed out")>0 || match($0, /all hosts for '\''[^']+'\'' have been failing/))) {
    is_defer=1
  }

  # Fail detection (permanent)
  # Zeilen mit "** rcpt ..." sind i. d. R. permanent; weitere Marker: "Unrouteable address", "rejected", "mailbox full" kann defer sein -> belassen
  is_perm_fail = 0
  if (is_fail) {
    is_perm_fail=1
  } else if (match($0, /(Unrouteable address|no such user|User unknown|550 )/)) {
    is_perm_fail=1
  }

  # Domainfilter (weicher): wir verwerfen die ID nicht, zeigen aber non-matching Zeilen trotzdem mit
  if (WANTDOM != "") {
    # keine harte Filterung, nur spätere Auswertung nutzt WANTDOM, falls nötig
    ;
  }

  # Extra Regex? (nur für Zeilenfilterung, ID bleibt)
  if (EXTRA != "" && $0 !~ EXTRA) {
    ; # wir speichern die Zeile dennoch im Voll-Log, sonst verliert man Kontext
  }

  # Counters
  if (is_success) deliveries[msgid]++
  if (is_defer)   defers[msgid]++
  if (is_perm_fail) fails[msgid]++

  # „have been failing …" Marker
  if (match($0, /all hosts for '\''[^']+'\'' have been failing/)) {
    fail_marker[msgid]=1
  }

  # Full log sammeln
  status[msgid]=(status[msgid] ? status[msgid] "\n" : "") ts " | " $0
}

END {
  # IDs einsammeln & sortieren
  n=0
  for (id in first_ts){ ids[++n]=id }
  for (i=2;i<=n;i++){
    key=ids[i]; j=i-1
    while (j>=1 && first_ts[ids[j]] > first_ts[key]) { ids[j+1]=ids[j]; j-- }
    ids[j+1]=key
  }

  if (n==0){
    print "Keine passenden Einträge im gewählten Zeitfenster gefunden."
    exit 0
  }

  sep="------------------------------------------------------------------------"
  for (k=1;k<=n;k++){
    id=ids[k]

    # FROM Nachfilter
    if (FROMREQ != "" && from_mismatch[id]==1) {
      if (from_addr[id] != FROMREQ && from_addr[id] != "") continue
    }

    d=(id in deliveries ? deliveries[id] : 0)
    e=(id in defers     ? defers[id]     : 0)
    f=(id in fails      ? fails[id]      : 0)

    # Status-Logik:
    # - d>0 && f==0 && e==0         -> OK
    # - d>0 && (f>0 || e>0)         -> PARTIAL
    # - d==0 && f>0                 -> FAIL
    # - d==0 && f==0 && e>0         -> DELAY
    # - sonst                       -> UNKNOWN
    status_txt="UNKNOWN"
    if (d>0 && f==0 && e==0) status_txt="OK"
    else if (d>0 && (f>0 || e>0)) status_txt="PARTIAL"
    else if (d==0 && f>0) status_txt="FAIL"
    else if (d==0 && f==0 && e>0) status_txt="DELAY"

    # Summary-Zeile
    # Rcpts sammeln
    out_rcpts=""
    if (id in rcpts){
      for (r in rcpts[id]) {
        out_rcpts = (out_rcpts=="" ? r : out_rcpts ", " r)
      }
    }
    printf "[%s] ID=%s | First=%s | From=%s | Rcpts=%s | Delivered=%d Deferred=%d Failed=%d\n",
      status_txt, id, first_ts[id], (from_addr[id] ? from_addr[id] : "(?)"), (out_rcpts=="" ? "(?)" : out_rcpts), d, e, f

    # Marker-Hinweis
    if (fail_marker[id]) {
      printf "  Note: DELIVERY STALL erkannt: '\''all hosts for domain failing (retry/backoff)'\''\n"
    }

    # Detailblock
    print sep
    printf "ID:        %s\n", id
    printf "First:     %s\n", first_ts[id]
    printf "Last:      %s\n", last_ts[id]
    printf "From:      %s\n", (from_addr[id] ? from_addr[id] : "(unbekannt)")
    printf "Rcpts:     %s\n", (out_rcpts=="" ? "(unbekannt)" : out_rcpts)
    printf "Status:    %s (Delivered=%d, Deferred=%d, Failed=%d)\n", status_txt, d, e, f
    print "Log:"
    print status[id]
    print sep
  }
}

# Built with the help of Claude Code