CASSANDRA-19703: Ensure prepared_statement INSERT timestamp precedes eviction DELETE #3917
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
Updates SystemKeyspace.writePreparedStatement to accept a timestamp associated with the Prepared creation time. Using this timestamp will ensure that an INSERT into system.prepared_statements will always precede the timestamp for the same Prepared in SystemKeyspace.removePreparedStatement. This is needed because Caffeine 2.9.2 may evict an entry as soon as it is inserted if the maximum weight of the cache is exceeded causing the DELETE to be executed before the INSERT. Additionally, any clusters currently experiencing a leaky system.prepared_statements table from this bug may struggle to bounce into a version with this fix as SystemKeyspace.loadPreparedPreparedStatements currently does not paginate the query to system.prepared_statements, causing heap OOMs. To fix this this patch adds pagingation at 5000 rows, which should allow nodes to come up and delete older prepared statements that may no longer be used as the cache fills up (which should happen immediately). Adds #testAsyncPstmtInvalidation which almost always reproduces the issue without this fix. Adds #testPreloadPreparedStatements to verify pagination behavior. This patch does not address the issue of Caffeine immediately evicting a prepared statement, however it will prevent the system.prepared_statements table from growing unbounded. For most users this should be adequate, as the cache should only be filled when there are erroneously many unique prepared statements. In such a case we can expect that clients will constantly prepare statements regardless of whether or not the cache is evicting statements.
tolbertam
force-pushed
the
CASSANDRA-19703-4.1
branch
from
1e639da to
9c1c572
Compare
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.
Updates SystemKeyspace.writePreparedStatement to accept a timestamp associated with the Prepared creation time. Using this timestamp will ensure that an INSERT into system.prepared_statements will always precede the timestamp for the same Prepared in SystemKeyspace.removePreparedStatement.
Additionally, any clusters currently experiencing a leaky system.prepared_statements table from this bug may struggle to bounce into a version with this fix as
SystemKeyspace.loadPreparedPreparedStatements currently does not paginate the query to system.prepared_statements, causing heap OOMs. To fix this this patch adds pagingation at 5000 rows, which should allow nodes to come up and delete older prepared statements that may no longer be used as
the cache fills up (which should happen immediately).
Adds #testAsyncPstmtInvalidation which almost always reproduces the issue without this fix.
Adds #testPreloadPreparedStatements to verify pagination behavior.
This patch does not address the issue of Caffeine immediately evicting a prepared statement, however it will prevent the
system.prepared_statements table from growing unbounded. For most users this should be adequate, as the cache should only be filled when there are erroneously many unique prepared statements. In such a case we can expect that clients will constantly prepare statements regardless of whether or not the cache is evicting statements.
CASSANDRA-19703