[bugfix](segmentload) should remove segment from segment cache if load segment failed (#41608) (#41660)

This commit is contained in:
yiguolei
2024-10-10 19:40:22 +08:00
committed by GitHub
parent 0fb42d3a48
commit f2ba1f2fb3
5 changed files with 60 additions and 33 deletions

View File

@ -312,25 +312,12 @@ Status Segment::_load_pk_bloom_filter() {
DCHECK(_tablet_schema->keys_type() == UNIQUE_KEYS);
DCHECK(_pk_index_meta != nullptr);
DCHECK(_pk_index_reader != nullptr);
auto status = [this]() {
return _load_pk_bf_once.call([this] {
RETURN_IF_ERROR(_pk_index_reader->parse_bf(_file_reader, *_pk_index_meta));
// _meta_mem_usage += _pk_index_reader->get_bf_memory_size();
return Status::OK();
});
}();
if (!status.ok()) {
remove_from_segment_cache();
}
return status;
}
void Segment::remove_from_segment_cache() const {
if (config::disable_segment_cache) {
return;
}
SegmentCache::CacheKey cache_key(_rowset_id, _segment_id);
SegmentLoader::instance()->erase_segment(cache_key);
return _load_pk_bf_once.call([this] {
RETURN_IF_ERROR(_pk_index_reader->parse_bf(_file_reader, *_pk_index_meta));
// _meta_mem_usage += _pk_index_reader->get_bf_memory_size();
return Status::OK();
});
}
Status Segment::load_pk_index_and_bf() {
@ -340,14 +327,6 @@ Status Segment::load_pk_index_and_bf() {
}
Status Segment::load_index() {
auto status = [this]() { return _load_index_impl(); }();
if (!status.ok()) {
remove_from_segment_cache();
}
return status;
}
Status Segment::_load_index_impl() {
return _load_index_once.call([this] {
if (_tablet_schema->keys_type() == UNIQUE_KEYS && _pk_index_meta != nullptr) {
_pk_index_reader = std::make_unique<PrimaryKeyIndexReader>();
@ -381,6 +360,32 @@ Status Segment::_load_index_impl() {
});
}
Status Segment::healthy_status() {
try {
if (_load_index_once.has_called()) {
RETURN_IF_ERROR(_load_index_once.stored_result());
}
if (_load_pk_bf_once.has_called()) {
RETURN_IF_ERROR(_load_pk_bf_once.stored_result());
}
if (_create_column_readers_once_call.has_called()) {
RETURN_IF_ERROR(_create_column_readers_once_call.stored_result());
}
if (_inverted_index_file_reader_open.has_called()) {
RETURN_IF_ERROR(_inverted_index_file_reader_open.stored_result());
}
// This status is set by running time, for example, if there is something wrong during read segment iterator.
return _healthy_status.status();
} catch (const doris::Exception& e) {
// If there is an exception during load_xxx, should not throw exception directly because
// the caller may not exception safe.
return e.to_status();
} catch (const std::exception& e) {
// The exception is not thrown by doris code.
return Status::InternalError("Unexcepted error during load segment: {}", e.what());
}
}
// Return the storage datatype of related column to field.
// Return nullptr meaning no such storage infomation for this column
vectorized::DataTypePtr Segment::get_data_type_of(const ColumnIdentifier& identifier,

View File

@ -135,6 +135,12 @@ public:
Status load_pk_index_and_bf();
void update_healthy_status(Status new_status) { _healthy_status.update(new_status); }
// The segment is loaded into SegmentCache and then will load indices, if there are something wrong
// during loading indices, should remove it from SegmentCache. If not, it will always report error during
// query. So we add a healthy status API, the caller should check the healhty status before using the segment.
Status healthy_status();
std::string min_key() {
DCHECK(_tablet_schema->keys_type() == UNIQUE_KEYS && _pk_index_meta != nullptr);
return _pk_index_meta->min_key();
@ -148,8 +154,6 @@ public:
int64_t meta_mem_usage() const { return _meta_mem_usage; }
void remove_from_segment_cache() const;
// Identify the column by unique id or path info
struct ColumnIdentifier {
int32_t unique_id = -1;
@ -212,7 +216,6 @@ private:
const SubcolumnColumnReaders::Node* root,
vectorized::DataTypePtr target_type_hint);
Status _load_index_impl();
Status _open_inverted_index();
Status _create_column_readers_once();
@ -222,6 +225,7 @@ private:
io::FileReaderSPtr _file_reader;
uint32_t _segment_id;
uint32_t _num_rows;
AtomicStatus _healthy_status;
// 1. Tracking memory use by segment meta data such as footer or index page.
// 2. Tracking memory use by segment column reader

View File

@ -269,8 +269,8 @@ SegmentIterator::SegmentIterator(std::shared_ptr<Segment> segment, SchemaSPtr sc
Status SegmentIterator::init(const StorageReadOptions& opts) {
auto status = _init_impl(opts);
if (!status.ok() && !config::disable_segment_cache) {
_segment->remove_from_segment_cache();
if (!status.ok()) {
_segment->update_healthy_status(status);
}
return status;
}
@ -1931,7 +1931,7 @@ Status SegmentIterator::next_batch(vectorized::Block* block) {
// if rows read by batch is 0, will return end of file, we should not remove segment cache in this situation.
if (!status.ok() && !status.is<END_OF_FILE>()) {
_segment->remove_from_segment_cache();
_segment->update_healthy_status(status);
}
return status;
}

View File

@ -59,8 +59,14 @@ Status SegmentLoader::load_segments(const BetaRowsetSharedPtr& rowset,
for (int64_t i = 0; i < rowset->num_segments(); i++) {
SegmentCache::CacheKey cache_key(rowset->rowset_id(), i);
if (_segment_cache->lookup(cache_key, cache_handle)) {
continue;
// Has to check the segment status here, because the segment in cache may has something wrong during
// load index or create column reader.
// Not merge this if logic with previous to make the logic more clear.
if (cache_handle->pop_unhealthy_segment() == nullptr) {
continue;
}
}
// If the segment is not healthy, then will create a new segment and will replace the unhealthy one in SegmentCache.
segment_v2::SegmentSharedPtr segment;
RETURN_IF_ERROR(rowset->load_segment(i, &segment));
if (need_load_pk_index_and_bf) {

View File

@ -162,6 +162,18 @@ public:
_init = true;
}
segment_v2::SegmentSharedPtr pop_unhealthy_segment() {
if (segments.empty()) {
return nullptr;
}
auto& last_segment = segments.back();
if (last_segment->healthy_status().ok()) {
return nullptr;
}
segments.pop_back();
return last_segment;
}
private:
std::vector<segment_v2::SegmentSharedPtr> segments;
bool _init {false};