antcolony

4.0 KB

1. #!/bin/bash
2. # cleanup.sh <from> <to> <since-ms> <session> — mission 026 (ticket antcolony#1): after a worker / controller / finalize
3. # session ended (any outcome), stop what it left behind: every process that
4. # * LISTENS (TCP) on a port of the session's range <from>-<to>, AND
5. # * STARTED after the session part began (<since-ms>, epoch ms; + 50 ms margin): start = now − (uptime − starttime)
6. # (/proc/uptime and /proc/<pid>/stat both count from boot — btime + starttime was up to 1 s off on Loreana), AND
7. # * belongs to this user, is not this script's own ancestor (the scheduler, its wrapper), AND
8. # * does not carry another colony session in its environment (COLONY_SESSION=<other>; the scheduler gives every
9. # session COLONY_SESSION=<its id>, which the worker's servers inherit).
10. # Never others. SIGTERM, up to 5 s, then SIGKILL. One line per process on stdout:
11. # stopped pid <p> · port <n> · started <ISO> · SIGTERM|SIGKILL · <command line>
12. # skipped pid <p> · port <n> · <why> · <command line> (or `skipped port <n> · no pid visible (another user)`)
13. # nothing listening on <from>-<to> started after the session (when nothing was found at all)
14. set -f # no globbing: `set -- $line` must not expand an ss address like `*:8781`
15. from=$1; to=$2; since_ms=$3; session=$4
16. case "$from$to$since_ms" in ''|*[!0-9]*) echo "error: usage: cleanup.sh <from> <to> <since-ms> <session>"; exit 0;; esac
17. me=$(id -u)
18. hz=$(getconf CLK_TCK)
19. # the age of a process = uptime − its starttime; wall clock and uptime read together once
20. now_ms=$(date +%s%3N)
21. up=$(cut -d' ' -f1 /proc/uptime)
22. up_ms=$(( ${up%.*} * 1000 + 10#${up#*.} * 10 ))
23. # ancestors of this script — never touched
24. anc=" "
25. p=$$
26. while [ -n "$p" ] && [ "$p" -gt 1 ] 2>/dev/null; do
27. anc="$anc$p "
28. s=$(cat "/proc/$p/stat" 2>/dev/null) || break
29. s=${s##*) }
30. set -- $s
31. p=$2
32. done
33. iso() { date -u -d "@$(( $1 / 1000 )).$(printf %03d $(( $1 % 1000 )))" +%Y-%m-%dT%H:%M:%S.%3NZ; }
34. # alive and not a zombie
35. alive() { local t; t=$(cat "/proc/$1/stat" 2>/dev/null) || return 1; t=${t##*) }; [ "${t%% *}" != Z ]; }
36. cmdof() { tr '\0' ' ' < "/proc/$1/cmdline" 2>/dev/null | cut -c1-160; }
37. declare -A seen seenport
38. n=0
39. while read -r line; do
40. [ -z "$line" ] && continue
41. set -- $line
42. local_addr=$4
43. port=${local_addr##*:}
44. case $port in ''|*[!0-9]*) continue;; esac
45. if [ "$port" -lt "$from" ] || [ "$port" -gt "$to" ]; then continue; fi
46. pids=$(printf '%s\n' "$line" | grep -o 'pid=[0-9]*' | cut -d= -f2 | sort -un)
47. if [ -z "$pids" ]; then
48. [ -n "${seenport[$port]}" ] && continue
49. seenport[$port]=1
50. echo "skipped port $port · no pid visible (another user's process)"
51. n=$((n + 1))
52. continue
53. fi
54. for pid in $pids; do
55. [ -n "${seen[$pid]}" ] && continue
56. seen[$pid]=1
57. n=$((n + 1))
58. [ -d "/proc/$pid" ] || { echo "skipped pid $pid · port $port · gone already"; continue; }
59. cmd=$(cmdof "$pid")
60. owner=$(stat -c %u "/proc/$pid" 2>/dev/null)
61. if [ "$owner" != "$me" ]; then echo "skipped pid $pid · port $port · another user ($owner) · $cmd"; continue; fi
62. case "$anc" in *" $pid "*) echo "skipped pid $pid · port $port · the scheduler itself · $cmd"; continue;; esac
63. s=$(cat "/proc/$pid/stat" 2>/dev/null) || { echo "skipped pid $pid · port $port · gone already"; continue; }
64. s=${s##*) }
65. set -- $s
66. st=${20}
67. start_ms=$(( now_ms - (up_ms - st * 1000 / hz) ))
68. if [ "$start_ms" -lt $(( since_ms + 50 )) ]; then echo "skipped pid $pid · port $port · started $(iso "$start_ms"), before the session · $cmd"; continue; fi
69. other=$(tr '\0' '\n' < "/proc/$pid/environ" 2>/dev/null | grep -m1 '^COLONY_SESSION=' | cut -d= -f2-)
70. if [ -n "$other" ] && [ "$other" != "$session" ]; then echo "skipped pid $pid · port $port · belongs to colony session $other · $cmd"; continue; fi
71. kill -TERM "$pid" 2>/dev/null
72. sig=SIGTERM
73. i=0
74. while alive "$pid" && [ $i -lt 50 ]; do sleep 0.1; i=$((i + 1)); done
75. if alive "$pid"; then kill -KILL "$pid" 2>/dev/null; sig=SIGKILL; fi
76. echo "stopped pid $pid · port $port · started $(iso "$start_ms") · $sig · $cmd"
77. done
78. done < <(ss -ltnpH 2>/dev/null)
79. [ "$n" -eq 0 ] && echo "nothing listening on $from-$to"
80. exit 0