Skip to content

fix(collector): isolate and drain complete undrained LOB segments #1156

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
proerror77 wants to merge 2 commits into
base: main
Choose a base branch
from
Open

fix(collector): isolate and drain complete undrained LOB segments #1156

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9eff487
fix(collector): isolate and drain complete undrained LOB segments
cursoragent Sep 12, 2026
3ef8c03
fix(collector): keep cleanup-only recovery jobs resumable
cursoragent Sep 12, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
44 changes: 42 additions & 2 deletions deployment/aliyun/host-rust-lob-recovery-queue.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -233,6 +233,43 @@ has_incomplete_parts() {
\) -print -quit) ]]
}

has_uploaded_cleanup_markers() {
local spool=$1
[[ -d $spool && ! -L $spool ]] || return 1
[[ -n $(find "$spool" -type f \( \
-name '*.uploaded-cleanup.json' -o \
-name '*.uploaded-cleanup.json.tmp' \
\) -print -quit) ]]
}

has_undrained_complete_segments() {
# Manifest/data/success only. Leftover uploaded-cleanup markers are the
# production archiver's post-upload residue, not undrained segments.
# Isolating them would send drain through --upload-only, which deletes the
# marker without refreshing last_success_at, then fails triplet verify and
# leaves the job unresumable.
local spool=$1
[[ -d $spool && ! -L $spool ]] || return 1
[[ -n $(find "$spool" -type f \( \
-name '*.manifest.json' -o \
-name '*.jsonl.zst' -o \
-name '*._SUCCESS' \
\) -print -quit) ]]
}

has_cleanup_marker_only_leftover() {
local spool=$1
has_uploaded_cleanup_markers "$spool" || return 1
has_incomplete_parts "$spool" && return 1
has_undrained_complete_segments "$spool" && return 1
return 0
}

needs_recovery_isolation() {
local spool=$1
has_incomplete_parts "$spool" || has_undrained_complete_segments "$spool"
}

segment_artifacts() {
local spool=$1
[[ -d $spool && ! -L $spool ]] || return 0
Expand Down Expand Up @@ -649,8 +686,8 @@ isolate_market() {
fail "canonical spool is missing; refusing recovery fallback: $CANONICAL_SPOOL"
fi
secure_directory "$CANONICAL_SPOOL" "$hft_uid" "$hft_gid"
isolation_phase_begin incomplete-scan
if ! has_incomplete_parts "$CANONICAL_SPOOL"; then
isolation_phase_begin recovery-scan
if ! needs_recovery_isolation "$CANONICAL_SPOOL"; then
isolation_phase_done
exit 0
fi
Expand Down Expand Up @@ -1285,6 +1322,9 @@ run_drain_job() {
"$release_binary" --recover-parts-only \
|| fail 'detached recovery of incomplete parts failed'
fi
if has_cleanup_marker_only_leftover "$running_dir"; then
fail 'detached spool has leftover uploaded-cleanup markers without local segment artifacts; refusing fresh-upload verify so the marker remains inspectable'
fi
# The uploader commits its own readback before removing local segments. An
# interrupted controller may therefore have only metadata left. Re-running
# an empty upload cannot refresh its timestamp and is not completion proof.
Expand Down
163 changes: 162 additions & 1 deletion deployment/aliyun/test-rust-lob-recovery-queue.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,17 @@ fi
grep -Fq "active V2 controller is required" "$RECOVERY"
grep -Fq 'monday.rust_lob_controller_release.v2' "$RECOVERY"
grep -Fq 'monday.rust_lob_controller_release.v2' "$RECOVERY"
grep -Fq 'needs_recovery_isolation' "$RECOVERY"
grep -Fq 'has_undrained_complete_segments' "$RECOVERY"
grep -Fq 'has_cleanup_marker_only_leftover' "$RECOVERY"
if grep -Fq 'if ! has_incomplete_parts "$CANONICAL_SPOOL"' "$RECOVERY"; then
printf 'recovery isolate still skips complete undrained segments\n' >&2
exit 1
fi
if awk '/^has_undrained_complete_segments\(\)/,/^}/' "$RECOVERY" | grep -Fq -- "-name '*.uploaded-cleanup"; then
printf 'undrained-complete still treats cleanup markers as segments\n' >&2
exit 1
fi
projection_contract='"$ACTIVE_CONTROLLER/deployment/binance-lob-archiver-production-$MARKET.env"'
resolved_contract='secure_regular_file "$installed_env" 0'
obsolete_contract='secure_regular_file "$ENV_FILE" 0'
Expand Down Expand Up @@ -542,4 +553,154 @@ expect_rejected missing-remote-object fixture_drain
unset FIXTURE_OSS_FAIL
[[ $(jq -r .result "$attempt/result.json") == failed ]]
[[ $(sha256sum "$EVIDENCE_ROOT/$RESUME_JOB_ID/result.json" | awk '{print $1}') == "$fixture_old_result_sha" ]]
printf 'Explicit recovery adoption, historical readback, mixed drain and interruption behavior passed\n'

# Isolate/drain must cover rotated complete undrained segments, not only
# incomplete parts. Drain of a complete-only spool uses --upload-only.
empty_spool="$fixture/empty-isolate-spool"
complete_spool="$fixture/complete-isolate-spool"
part_spool="$fixture/part-isolate-spool"
mkdir -p "$empty_spool" "$complete_spool" "$part_spool"
printf '{}\n' >"$empty_spool/upload-status.json"
printf '{}\n' >"$empty_spool/health.json"
: >"$empty_spool/.binance-lob-archiver.lock"
if has_incomplete_parts "$empty_spool" || has_undrained_complete_segments "$empty_spool" \
|| needs_recovery_isolation "$empty_spool"; then
printf 'empty spool was treated as recovery work\n' >&2
exit 1
fi
printf 'sealed\n' >"$complete_spool/part-1.jsonl.zst"
printf '{}\n' >"$complete_spool/part-1.jsonl.zst.manifest.json"
printf 'ok\n' >"$complete_spool/part-1.jsonl.zst._SUCCESS"
has_undrained_complete_segments "$complete_spool"
needs_recovery_isolation "$complete_spool"
if has_incomplete_parts "$complete_spool"; then
printf 'complete segment was classified as an incomplete part\n' >&2
exit 1
fi
printf 'raw\n' >"$part_spool/part-2.jsonl.part"
has_incomplete_parts "$part_spool"
needs_recovery_isolation "$part_spool"
if has_undrained_complete_segments "$part_spool"; then
printf 'incomplete part was classified as a complete segment\n' >&2
exit 1
fi
cleanup_spool="$fixture/cleanup-isolate-spool"
mkdir -p "$cleanup_spool"
printf '%s\n' '{"schema":"monday.binance_lob.uploaded_cleanup.v1","data":"part-1.jsonl.zst","manifest":"part-1.jsonl.zst.manifest.json","success":"part-1.jsonl.zst._SUCCESS"}' \
>"$cleanup_spool/part-1.jsonl.zst.manifest.json.uploaded-cleanup.json"
has_uploaded_cleanup_markers "$cleanup_spool"
has_cleanup_marker_only_leftover "$cleanup_spool"
if has_undrained_complete_segments "$cleanup_spool" || needs_recovery_isolation "$cleanup_spool"; then
printf 'cleanup-marker-only spool was treated as recovery work\n' >&2
exit 1
fi

rm -f "$fixture/payload.calls" "$fixture/uploaded-files"
fixture_job 108 ready
printf 'already sealed\n' >"$fixture_job_dir/part-complete.jsonl.zst"
printf '{}\n' >"$fixture_job_dir/part-complete.jsonl.zst.manifest.json"
printf 'success\n' >"$fixture_job_dir/part-complete.jsonl.zst._SUCCESS"
fixture_resume >/dev/null
attempt=$(fixture_attempt)
fixture_drain >/dev/null
[[ -d $attempt/spool.done && $(jq -r .result "$attempt/result.json") == passed ]]
grep -Fq -- '--upload-only' "$fixture/payload.calls"
if grep -Fq -- '--recover-parts-only' "$fixture/payload.calls"; then
printf 'complete-only drain invoked recover-parts\n' >&2
exit 1
fi
grep -Fq 'part-complete.jsonl.zst' "$fixture/uploaded-files"

rm -f "$fixture/payload.calls" "$fixture/uploaded-files"
fixture_job 109 ready
printf '%s\n' '{"schema":"monday.binance_lob.uploaded_cleanup.v1","data":"part-cleanup.jsonl.zst","manifest":"part-cleanup.jsonl.zst.manifest.json","success":"part-cleanup.jsonl.zst._SUCCESS"}' \
>"$fixture_job_dir/part-cleanup.jsonl.zst.manifest.json.uploaded-cleanup.json"
fixture_resume >/dev/null
attempt=$(fixture_attempt)
expect_rejected leftover-cleanup-markers fixture_drain
grep -Fq 'leftover uploaded-cleanup markers' "$fixture/rejected.log"
[[ -d $QUEUE_MARKET_ROOT/$RESUME_JOB_ID.failed ]]
[[ -f $QUEUE_MARKET_ROOT/$RESUME_JOB_ID.failed/part-cleanup.jsonl.zst.manifest.json.uploaded-cleanup.json ]]
[[ $(jq -r .result "$attempt/result.json") == failed ]]
if [[ -e $fixture/payload.calls ]] && grep -Fq -- '--upload-only' "$fixture/payload.calls"; then
printf 'cleanup-marker-only drain entered fresh-upload verify\n' >&2
exit 1
fi
if [[ -e $fixture/payload.calls ]] && grep -Fq -- '--recover-parts-only' "$fixture/payload.calls"; then
printf 'cleanup-marker-only drain invoked recover-parts\n' >&2
exit 1
fi
fixture_drain >/dev/null
[[ -d $QUEUE_MARKET_ROOT/$RESUME_JOB_ID.failed ]]
[[ ! -d $QUEUE_MARKET_ROOT/$RESUME_JOB_ID.ready ]]

saved_canonical=$CANONICAL_SPOOL
isolate_spool="$fixture/host/data/monday/spool/binance-lob/spot"
rm -rf -- "$isolate_spool"
mkdir -p "$isolate_spool"
printf '{}\n' >"$isolate_spool/upload-status.json"
: >"$isolate_spool/.binance-lob-archiver.lock"
chmod 0640 "$isolate_spool/.binance-lob-archiver.lock" "$isolate_spool/upload-status.json"
ready_before=$(find "$QUEUE_MARKET_ROOT" -mindepth 1 -maxdepth 1 -type d -name '*.ready' -print | wc -l)
fixture_isolate() {
(
CURRENT_ACTION=isolate
# shellcheck disable=SC2317,SC2329 # Nested stub used by run_isolate in this subshell.
install() {
local mode=0750 dest src
while (($#)); do
case "$1" in
-d) shift ;;
-m) mode=$2; shift 2 ;;
-o|-g) shift 2 ;;
--) shift; break ;;
*) break ;;
esac
done
dest=${*: -1}
if (($# <= 1)); then
mkdir -p "$dest"
chmod "$mode" "$dest" 2>/dev/null || true
else
src=${*: -2:1}
cp -- "$src" "$dest"
chmod "$mode" "$dest" 2>/dev/null || true
fi
}
queue_lock
run_isolate
)
}
if ! fixture_isolate >/dev/null; then
printf 'empty canonical spool isolate failed\n' >&2
exit 1
fi
ready_after=$(find "$QUEUE_MARKET_ROOT" -mindepth 1 -maxdepth 1 -type d -name '*.ready' -print | wc -l)
[[ $ready_before == "$ready_after" ]] \
|| { printf 'empty spool isolate queued a job\n' >&2; exit 1; }

printf '%s\n' '{"schema":"monday.binance_lob.uploaded_cleanup.v1","data":"part-cleanup.jsonl.zst","manifest":"part-cleanup.jsonl.zst.manifest.json","success":"part-cleanup.jsonl.zst._SUCCESS"}' \
>"$isolate_spool/part-cleanup.jsonl.zst.manifest.json.uploaded-cleanup.json"
if ! fixture_isolate >/dev/null; then
printf 'cleanup-marker-only isolate failed\n' >&2
exit 1
fi
ready_after_cleanup=$(find "$QUEUE_MARKET_ROOT" -mindepth 1 -maxdepth 1 -type d -name '*.ready' -print | wc -l)
[[ $ready_before == "$ready_after_cleanup" ]] \
|| { printf 'cleanup-marker-only isolate queued a job\n' >&2; exit 1; }
[[ -f $isolate_spool/part-cleanup.jsonl.zst.manifest.json.uploaded-cleanup.json ]]
rm -f -- "$isolate_spool/part-cleanup.jsonl.zst.manifest.json.uploaded-cleanup.json"

printf 'sealed\n' >"$isolate_spool/part-pending.jsonl.zst"
printf '{}\n' >"$isolate_spool/part-pending.jsonl.zst.manifest.json"
printf 'ok\n' >"$isolate_spool/part-pending.jsonl.zst._SUCCESS"
fixture_isolate >/dev/null
ready_dir=$(find "$QUEUE_MARKET_ROOT" -mindepth 1 -maxdepth 1 -type d -name '*.ready' -print \
| while IFS= read -r dir; do
[[ -f $dir/part-pending.jsonl.zst ]] && printf '%s\n' "$dir"
done | sed -n '1p')
[[ -n $ready_dir ]] || { printf 'complete undrained isolate did not queue a job\n' >&2; exit 1; }
[[ -f $ready_dir/part-pending.jsonl.zst && -f $ready_dir/part-pending.jsonl.zst.manifest.json ]]
[[ -f $isolate_spool/upload-status.json && ! -e $isolate_spool/part-pending.jsonl.zst ]]
CANONICAL_SPOOL=$saved_canonical
printf 'Explicit recovery adoption, historical readback, mixed drain and complete-segment isolation passed\n'
Loading