[CI crash] Incorrect destruction of MergeTreeDataPartCompact caches #94661
Description
Stack trace details
The sipHash64(st.trace_full) is 17987466783627542056
The trace is from the master or release branch: True
The query for CIDB to compare the trace with the known one:
WITH
(
SELECT groupArrayDistinct(cleanStackTrace(trace_full) AS trace) FROM default.stack_traces
WHERE sipHash64(trace) IN (17987466783627542056, {ANOTHER_TRACE_HASH}) -- FIXME: replace with the known hash
) AS traces,
1.97 AS alpha,
stack_frame_weights AS (
WITH
(
SELECT count()
FROM default.stack_traces
FINAL
) AS total,
2.0 AS beta,
3.7 AS gamma
SELECT
arrayJoin(cleanStackTrace(trace_full)) AS frame,
countDistinct(trace_full) AS count,
log(total / count) AS IDF,
sigmoid(beta * (IDF - gamma)) AS weight
FROM default.stack_traces
FINAL
GROUP BY frame
),
(SELECT groupArray(weight) AS w, groupArray(frame) AS f FROM stack_frame_weights) AS weights,
(trace -> arrayMap((_frame, pos) -> (pow(pos, -alpha) * arrayFirst(w, f -> (f = _frame), weights.w, weights.f)), trace, arrayEnumerate(trace))) AS get_trace_weights,
(arr -> arrayStringConcat(arr, '\n')) AS joinArr
SELECT arraySimilarity(traces[1], traces[2], get_trace_weights(traces[1]) AS weights1, get_trace_weights(traces[2]) AS weights2) AS similarity,
arrayLevenshteinDistanceWeighted(traces[1], traces[2], weights1, weights2),
joinArr(traces[1]), joinArr(traces[2]), joinArr(weights1), joinArr(weights2)
The following new stack trace from CI Logs system.crash_log found:
DB::MergeTreeData::getPrimaryIndexCache() const
DB::IMergeTreeDataPart::clearCaches()
DB::IMergeTreeDataPart::removeIfNeeded()
DB::MergeTreeDataPartCompact::~MergeTreeDataPartCompact()
DB::RangesInDataPart::~RangesInDataPart()
DB::MergeTreeReadPoolBase::~MergeTreeReadPoolBase()
DB::MergeTreeReadPool::~MergeTreeReadPool()
DB::MergeTreeSelectProcessor::~MergeTreeSelectProcessor()
DB::MergeTreeSource::~MergeTreeSource()
std::__shared_ptr_emplace<std::vector<std::shared_ptr<DB::IProcessor>, std::allocator<std::shared_ptr<DB::IProcessor>>>, std::allocator<std::vector<std::shared_ptr<DB::IProcessor>, std::allocator<std::shared_ptr<DB::IProcessor>>>>>::__on_zero_shared()
DB::QueryPipeline::~QueryPipeline()
DB::QueryPipeline::reset()
DB::TCPHandler::runImpl()
DB::TCPHandler::run()
Poco::Net::TCPServerConnection::start()
Poco::Net::TCPServerDispatcher::run()
Poco::PooledThread::run()
Poco::ThreadImpl::runnableEntry(void*)
Possible causes:
- Improper cleanup of cache objects during destruction
- Double deletion or invalid reference to cache data
- Race condition during cache destruction in multi-threaded environment
- Cache not properly invalidated before destruction
The stack trace appeared in the following checks: