mirror of
https://github.com/pgadmin-org/pgadmin4.git
synced 2026-10-29 07:50:04 -05:00
Add a spike workflow to diagnose the macOS DMG detach failure
Three Intel appbundle runs have now failed identically, always exactly 120 seconds after create-dmg starts unmounting, with the bundle otherwise complete and verified; arm64 has passed four times through the same code, and disabling Spotlight changed nothing. Reproducing it through `make appbundle` costs forty minutes a go, nearly all of it compiling cryptography on Intel, which is a miserable way to test a hypothesis. So this reproduces the failing step directly, in a couple of minutes, and answers three questions at once: whether plain hdiutil fails or only create-dmg does, whether it depends on image size, and whether it is specific to the macos-15-intel image. arm64 is included as the control that is known to work. When something fails it captures evidence rather than leaving us guessing: hdiutil's own view of what is attached, lsof against the volume, the processes that typically hold one, the diskarbitrationd and hdiutil unified log for the preceding few minutes, and whether a forced detach succeeds where the polite one timed out. That last one should be the most telling, since it distinguishes something holding a reference from DiskArbitration itself not answering. The mdutil step added to check-macos-build.yml stays for now so the two runs remain comparable; it should come out once we know the real cause.
This commit is contained in:
Dave Page committed
1 parent
68a5f7781f
commit
d5c71d7063
1 file changed
+200
@@ -0,0 +1,200 @@
|
||||
# Throwaway diagnostic spike. NOT part of CI.
|
||||
#
|
||||
# `hdiutil detach` fails deterministically on macos-15-intel at the end of
|
||||
# create-dmg, always exactly 120 seconds after the unmount starts:
|
||||
#
|
||||
# Unmounting disk image...
|
||||
# hdiutil: detach: timeout for DiskArbitration expired
|
||||
# hdiutil: detach: drive not detached
|
||||
#
|
||||
# Three appbundle runs, three identical failures, while arm64 passed four times
|
||||
# through the same code. Disabling Spotlight made no difference.
|
||||
#
|
||||
# Reproducing that through `make appbundle` costs 40 minutes a go, most of it
|
||||
# compiling cryptography, so this reproduces it directly instead and takes a
|
||||
# couple of minutes. It answers three questions at once:
|
||||
#
|
||||
# 1. Is it create-dmg at all, or plain hdiutil? (test A vs test B)
|
||||
# 2. Does it depend on image size? (test A runs small and large)
|
||||
# 3. Is it the runner image? (the matrix includes macos-26-intel, with
|
||||
# arm64 as the control that is known to work)
|
||||
#
|
||||
# and when something does fail, it captures who is holding the volume rather
|
||||
# than leaving us to guess: hdiutil's own view, lsof, the diskarbitrationd
|
||||
# log, and whether a forced detach succeeds where the polite one did not.
|
||||
#
|
||||
# Delete once the cause is understood.
|
||||
|
||||
name: Spike - macOS DMG detach diagnostics
|
||||
|
||||
on:
|
||||
workflow_dispatch:
|
||||
|
||||
permissions:
|
||||
contents: read
|
||||
|
||||
jobs:
|
||||
dmg:
|
||||
runs-on: ${{ matrix.runner }}
|
||||
timeout-minutes: 45
|
||||
|
||||
strategy:
|
||||
fail-fast: false
|
||||
matrix:
|
||||
runner:
|
||||
- 'macos-15-intel'
|
||||
- 'macos-26-intel'
|
||||
- 'macos-15'
|
||||
|
||||
steps:
|
||||
- uses: actions/checkout@v7
|
||||
|
||||
- name: Record the runner specification
|
||||
run: |
|
||||
sw_vers
|
||||
echo "Architecture: $(uname -m)"
|
||||
echo "CPUs: $(sysctl -n hw.logicalcpu)"
|
||||
echo "RAM: $(( $(sysctl -n hw.memsize) / 1024 / 1024 )) MB"
|
||||
df -h /
|
||||
|
||||
# Written once and reused by every test below, so a failure anywhere
|
||||
# produces the same evidence.
|
||||
- name: Install the diagnostic helper
|
||||
run: |
|
||||
cat > "${RUNNER_TEMP}/diagnose.sh" <<'EOF'
|
||||
#!/bin/bash
|
||||
# $1 = mount point (may be empty), $2 = device node (may be empty)
|
||||
MOUNT="$1"
|
||||
DEV="$2"
|
||||
echo "======== DIAGNOSTICS ========"
|
||||
|
||||
echo "---- hdiutil info ----"
|
||||
hdiutil info || true
|
||||
|
||||
echo "---- mounted volumes ----"
|
||||
ls -l /Volumes/ || true
|
||||
mount || true
|
||||
|
||||
echo "---- lsof against the volume ----"
|
||||
if [ -n "${MOUNT}" ]; then
|
||||
lsof "${MOUNT}" 2>/dev/null || true
|
||||
lsof +D "${MOUNT}" 2>/dev/null || true
|
||||
fi
|
||||
lsof 2>/dev/null | grep -F '/Volumes/' || echo "(no lsof rows mentioning /Volumes/)"
|
||||
|
||||
echo "---- processes that tend to hold volumes ----"
|
||||
ps aux | grep -E 'mds|mds_stores|fseventsd|diskarbitrationd|Spotlight|antivirus' \
|
||||
| grep -v grep || echo "(none)"
|
||||
|
||||
echo "---- diskarbitrationd log, last 3 minutes ----"
|
||||
sudo log show --last 3m --predicate \
|
||||
'process == "diskarbitrationd" OR process == "hdiutil" OR process == "diskimages-helper"' \
|
||||
--style compact 2>/dev/null | tail -60 || echo "(log show unavailable)"
|
||||
|
||||
echo "---- does a forced detach work where the polite one did not? ----"
|
||||
if [ -n "${DEV}" ]; then
|
||||
time hdiutil detach "${DEV}" -force && echo "FORCED DETACH: SUCCEEDED" \
|
||||
|| echo "FORCED DETACH: ALSO FAILED"
|
||||
fi
|
||||
echo "======== END DIAGNOSTICS ========"
|
||||
EOF
|
||||
chmod +x "${RUNNER_TEMP}/diagnose.sh"
|
||||
|
||||
# Test A: no create-dmg, no pgAdmin, just hdiutil. If this fails, the
|
||||
# problem is the runner's disk image handling and create-dmg is a
|
||||
# bystander. Two sizes, because a 120-second timeout smells like it
|
||||
# could scale with how much there is to flush.
|
||||
- name: 'Test A: bare hdiutil attach/detach'
|
||||
id: test_a
|
||||
continue-on-error: true
|
||||
run: |
|
||||
set -o pipefail
|
||||
for SIZE in 100m 1500m; do
|
||||
echo "############ bare hdiutil, ${SIZE} ############"
|
||||
IMG="${RUNNER_TEMP}/bare-${SIZE}.dmg"
|
||||
rm -f "${IMG}"
|
||||
hdiutil create -size "${SIZE}" -fs HFS+ -volname "BareTest" "${IMG}"
|
||||
|
||||
DEV=$(hdiutil attach "${IMG}" -nobrowse -noverify -noautoopen \
|
||||
| grep -E '^/dev/' | head -1 | awk '{print $1}')
|
||||
MOUNT=$(mount | grep -F "${DEV}" | sed -E 's/.* on (.*) \(.*/\1/')
|
||||
echo "device=${DEV} mount=${MOUNT}"
|
||||
|
||||
# Put something in it, so the detach has buffers to flush.
|
||||
dd if=/dev/urandom of="${MOUNT}/payload" bs=1m count=50 2>/dev/null
|
||||
sync
|
||||
|
||||
echo "---- detaching ----"
|
||||
START=$(date +%s)
|
||||
if hdiutil detach "${DEV}"; then
|
||||
echo "DETACH OK after $(( $(date +%s) - START ))s (${SIZE})"
|
||||
else
|
||||
echo "DETACH FAILED after $(( $(date +%s) - START ))s (${SIZE})"
|
||||
"${RUNNER_TEMP}/diagnose.sh" "${MOUNT}" "${DEV}"
|
||||
exit 1
|
||||
fi
|
||||
done
|
||||
|
||||
# Test B: create-dmg with pgAdmin's real flags and real assets, over a
|
||||
# payload of roughly the size of the actual bundle. This is the failing
|
||||
# path, minus the 40 minutes of build in front of it.
|
||||
- name: 'Test B: create-dmg with pgAdmin flags'
|
||||
id: test_b
|
||||
continue-on-error: true
|
||||
run: |
|
||||
set -o pipefail
|
||||
git clone --depth 1 https://github.com/create-dmg/create-dmg.git "${RUNNER_TEMP}/create-dmg"
|
||||
echo "create-dmg at $(git -C "${RUNNER_TEMP}/create-dmg" rev-parse --short HEAD)"
|
||||
|
||||
# A stand-in for the .app: a few hundred MB of mixed compressible and
|
||||
# incompressible data, since the real DMG is UDBZ and the bundle is
|
||||
# mostly a Python framework plus Electron.
|
||||
APP="${RUNNER_TEMP}/pgAdmin 4.app"
|
||||
mkdir -p "${APP}/Contents/MacOS"
|
||||
dd if=/dev/urandom of="${APP}/Contents/MacOS/blob" bs=1m count=400 2>/dev/null
|
||||
for i in $(seq 1 200); do
|
||||
cp "${GITHUB_WORKSPACE}/README.md" "${APP}/Contents/text-${i}.md"
|
||||
done
|
||||
du -sh "${APP}"
|
||||
|
||||
mkdir -p "${RUNNER_TEMP}/out"
|
||||
START=$(date +%s)
|
||||
set +e
|
||||
"${RUNNER_TEMP}/create-dmg/create-dmg" \
|
||||
--volname "pgAdmin 4" \
|
||||
--volicon "${GITHUB_WORKSPACE}/pkg/mac/dmg-icon.icns" \
|
||||
--eula "${GITHUB_WORKSPACE}/pkg/mac/licence.rtf" \
|
||||
--background "${GITHUB_WORKSPACE}/pkg/mac/dmg-background.png" \
|
||||
--app-drop-link 600 220 \
|
||||
--icon "pgAdmin 4.app" 200 220 \
|
||||
--window-pos 200 120 \
|
||||
--window-size 800 400 \
|
||||
--hide-extension "pgAdmin 4.app" \
|
||||
--add-file .DS_Store "${GITHUB_WORKSPACE}/pkg/mac/dmg.DS_Store" 5 5 \
|
||||
--format UDBZ \
|
||||
--skip-jenkins \
|
||||
--no-internet-enable \
|
||||
"${RUNNER_TEMP}/out/test.dmg" \
|
||||
"${APP}"
|
||||
RC=$?
|
||||
set -e
|
||||
echo "create-dmg exited ${RC} after $(( $(date +%s) - START ))s"
|
||||
|
||||
if [ ${RC} -ne 0 ]; then
|
||||
# create-dmg leaves the volume mounted when the detach fails, so
|
||||
# find it rather than assuming a name.
|
||||
MOUNT=$(ls -d /Volumes/dmg.* 2>/dev/null | head -1)
|
||||
DEV=$(mount | grep -F "${MOUNT}" | awk '{print $1}')
|
||||
echo "leftover mount=${MOUNT} device=${DEV}"
|
||||
"${RUNNER_TEMP}/diagnose.sh" "${MOUNT}" "${DEV}"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
ls -lh "${RUNNER_TEMP}/out/"
|
||||
|
||||
- name: Summarise
|
||||
if: always()
|
||||
run: |
|
||||
echo "runner: ${{ matrix.runner }}"
|
||||
echo "test A (bare hdiutil): ${{ steps.test_a.outcome }}"
|
||||
echo "test B (create-dmg): ${{ steps.test_b.outcome }}"
|
||||
Reference in new issue
Block a user