shishantbiswas/coder
1
0
Fork 0
mirror of https://github.com/coder/coder.git synced 2026-06-02 20:48:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(scripts/develop.sh): handle SIGHUP and prevent SIGPIPE during shutdown (#22994)

Browse Source 
When SSH disconnects, the output reader subshells fail writing to the
dead terminal (EIO) and exit due to set -e. This breaks the pipe to
the server, which receives SIGPIPE and dies before graceful shutdown
can stop embedded PostgreSQL.

Disable errexit in readers so they survive terminal death, add HUP
traps so the script handles SSH disconnect, and shield children from
direct SIGHUP via SIG_IGN before exec (Go resets this to caught once
signal.Notify registers). Also make fatal() exit immediately instead
of relying on async signal delivery, and remove the broken process
group kill (ppid was never a PGID).
This commit is contained in:
Mathias Fredriksson
2026-03-12 16:09:30 +02:00
committed by GitHub
parent 3aada03f52
commit 2bb483b425
1 changed files with 26 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/develop.sh
+26 -14
View File
@@ -157,24 +157,23 @@ pids=()
exit_cleanup() {
set +e
# Set empty interrupt handler so cleanup isn't interrupted.
trap '' INT TERM
# HUP is included in case SSH drops while cleanup is already
# in progress from another signal.
trap '' INT TERM HUP
# Remove exit trap to avoid infinite loop.
trap - EXIT
# Send interrupts to the processes we started. Note that we do not
# (yet) want to send a kill signal to the entire process group as
# this can halt processes started by graceful shutdown.
# Send INT for graceful shutdown. On SIGHUP, the Go server
# re-registers via signal.Notify and handles it, but other
# commands may not. INT covers all cases uniformly.
kill -INT "${pids[@]}" >/dev/null 2>&1
# Use the hammer if things take too long.
{ sleep 5 && kill -TERM "${pids[@]}" >/dev/null 2>&1; } &
# Use the hammer if things take too long. Stdout/stderr are
# closed so the background job can't hold the shell open.
{ sleep 15 && kill -TERM "${pids[@]}"; } >/dev/null 2>&1 &
# Wait for all children to exit (this can be aborted by hammer).
wait_cmds
# Just in case, send termination to the entire process group
# in case the children left something behind.
kill -TERM -"${ppid}" >/dev/null 2>&1
exit 1
}
start_cmd() {
@@ -184,9 +183,18 @@ start_cmd() {
echo "== CMD: $*" >&2
FORCE_COLOR=1 "$@" > >(
# Ignore interrupt, read will keep reading until stdin is gone.
trap '' INT
# Shield the command from direct SIGHUP via SIG_IGN, which
# is inherited across exec. Go's signal.Notify resets it to
# caught once registered, enabling graceful shutdown.
(
trap '' HUP
FORCE_COLOR=1 exec "$@"
) > >(
# Keep draining output until the command's stdout closes.
# Errexit is off so EIO from a dead terminal doesn't kill
# this reader and break the pipe (causing SIGPIPE).
set +e
trap '' INT HUP
while read -r line; do
if [[ $prefix == date ]]; then
@@ -207,6 +215,10 @@ wait_cmds() {
fatal() {
echo "== FAIL: $*" >&2
kill -INT $ppid >/dev/null 2>&1
# Exit immediately so the calling context (ERR trap or
# background job) doesn't continue executing commands
# while cleanup is in progress.
exit 1
}
# This is a way to run multiple processes in parallel, and have Ctrl-C work correctly
@@ -216,7 +228,7 @@ fatal() {
ppid=$BASHPID
# If something goes wrong, just bail and tear everything down
# rather than leaving things in an inconsistent state.
trap 'exit_cleanup' INT TERM EXIT
trap 'exit_cleanup' INT TERM HUP EXIT
trap 'fatal "Script encountered an error"' ERR
cdroot