PS-9391 Fixes replication break error because of HASH_SCAN #5528
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
percona-ysorokin
requested changes
Jan 6, 2025
| @@ -2898,14 +2898,34 @@ class Rows_log_event : public virtual binary_log::Rows_event, public Log_event { | |||
| Key_compare(KEY **ki = nullptr) : m_key_info(ki) {} | |||
| bool operator()(uchar *k1, uchar *k2) const { | |||
| return key_cmp2((*m_key_info)->key_part, k1, (*m_key_info)->key_length, | |||
| k2, (*m_key_info)->key_length) < 0; | |||
| k2, (*m_key_info)->key_length) == 0; | |||
There was a problem hiding this comment.
Rename this helper class to Key_equal.
Compare may be misleading as it may mean comparator and in your scenario this is no longer the case. All you do now is checking for equality.
| class Key_hash { | ||
| public: | ||
| Key_hash(KEY **ki = nullptr) : m_key_info(ki) {} | ||
| size_t operator()(uchar* ptr) const { |
There was a problem hiding this comment.
May be just create a std::string_view from ptr and (*m_key_info)->key_length and calculate a hash of this string_view using standard hash function.
std::string_view sv{ptr, (*m_key_info)->key_length};
return std::hash(sv);There was a problem hiding this comment.
Thanks for the suggestion. Done.
sql/log_event.cc
Outdated
| @@ -9142,9 +9141,10 @@ int Rows_log_event::add_key_to_distinct_keyset() { | |||
| DBUG_TRACE; | |||
| assert(m_key_index < MAX_KEY); | |||
| key_copy(m_distinct_key_spare_buf, m_table->record[0], m_key_info, 0); | |||
| std::pair<std::set<uchar *, Key_compare>::iterator, bool> ret = | |||
| std::pair<std::unordered_set<uchar *, Key_hash, Key_compare>::iterator, bool> ret = | |||
There was a problem hiding this comment.
May be just auto?
| @@ -9088,14 +9090,11 @@ int Rows_log_event::open_record_scan() { | |||
|
|
|||
| if (m_key_index < MAX_KEY) { | |||
| if (m_rows_lookup_algorithm == ROW_LOOKUP_HASH_SCAN) { | |||
| /* initialize the iterator over the list of distinct keys that we have */ | |||
| m_itr = m_distinct_keys.begin(); | |||
There was a problem hiding this comment.
Shouldn't we change this to m_distinct_key_id = 0?
There was a problem hiding this comment.
m_distinct_key_idx is already set to 0 when it is declared.
sql/log_event.cc
Outdated
| @@ -9068,8 +9068,10 @@ int Rows_log_event::next_record_scan(bool first_read) { | |||
| if ((error = table->file->ha_index_read_map( | |||
| table->record[0], m_key, HA_WHOLE_KEY, HA_READ_KEY_EXACT))) { | |||
| DBUG_PRINT("info", ("no record matching the key found in the table")); | |||
| if (!is_trx_retryable_upon_engine_error(error)) | |||
| if (!is_trx_retryable_upon_engine_error(error)) { | |||
| sleep(10); | |||
There was a problem hiding this comment.
Is this intentional or just for debugging?
There was a problem hiding this comment.
It was just for debugging. Removed it.
https://perconadev.atlassian.net/browse/PS-9391 Problem ======= When a replica's slave_rows_search_algorithms is set to HASH_SCAN, the replication may break with HA_ERR_KEY_NOT_FOUND. Analysis ======== When a replica's slave_rows_search_algorithms is set to HASH_SCAN, it prepares a unique key list for all the rows in a particular Row_event. The same unique key list will later be used to retrieve all tuples associated to each key in the list from storage engine. In the case of multiple updates targeted at the same row like how it is shown in the testcase, it may happen that this unique key list filled with entries which don't exist yet in the table. This is a problem when there is an intermediate update which changes the value of the index column to a lesser value than the original entry and that changed value is used in another update as shown in the second part of the testcase. It is an issue because the unique key list is a std::set which internally sorts it's entries. When this sorting happens, the first entry of the list could potentially be a value which doesn't exist in the table and the when it is searched in next_record_scan() method, it fails returning HA_ERR_KEY_NOT_FOUND error. Solution ======== Instead of using std::set to store the distinct keys, a combination of unordered_set and a list is used to preserve the original order of updates and avoid duplicates at the same time which prevents the side effects of sorting.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
https://perconadev.atlassian.net/browse/PS-9391
Problem
When a replica's slave_rows_search_algorithms is set to HASH_SCAN, the
replication may break with HA_ERR_KEY_NOT_FOUND.
Analysis
When a replica's slave_rows_search_algorithms is set to HASH_SCAN,
it prepares a unique key list for all the rows in a particular Row_event.
The same unique key list will later be used to retrieve all tuples associated
to each key in the list from storage engine. In the case of multiple updates
targeted at the same row like how it is shown in the testcase, it may happen
that this unique key list filled with entries which don't exist yet in the table.
This is a problem when there is an intermediate update which changes the value
of the index column to a lesser value than the original entry and that changed
value is used in another update as shown in the second part of the testcase.
It is an issue because the unique key list is a std::set which internally
sorts it's entries. When this sorting happens, the first entry of the list
could potentially be a value which doesn't exist in the table and the when
it is searched in next_record_scan() method, it fails returning
HA_ERR_KEY_NOT_FOUND error.
Solution
Instead of using std::set to store the distinct keys, a combination of
unordered_set and a list is used to preserve the original order of
updates and avoid duplicates at the same time which prevents the side
effects of sorting.