Fix estimateRowOverhead in Memtable: ByteBuffer in Partition key was not counted
Fix non-counted size of "columns" field in Columns.unsharedHeapSize
BTree.sizeOfStructureOnHeap wrongly reported a non-zero size for EMPTY_LEAF shared constant value
Fix MemtableSizeTest flakeness by not measuring of allocator and ColumnFamilyStore/Keyspace
Patch by Dmitry Konstantinov; reviewed by Branimir Lambov, Stefan Miklosovic for CASSANDRA-17298
patch by Caleb Rackliffe; reviewed by Zhao Yang, Brandon Williams, and Marcus Eriksson for CASSANDRA-19704
Co-authored-by: Marcus Eriksson <marcuse@apache.org>
Co-authored-by: Caleb Rackliffe <calebrackliffe@gmail.com>
nodetool import requires keyspace and table on the command line to import SSTables to. This
was not always working as specifying keypace and table while having SSTables located in the
directory structure which differed on keyspace and table (dir and parent dir) just
ignored them.
patch by Stefan Miklosovic; reviewed by Brandon Williams for CASSANDRA-19401
patch by Andrés de la Peña; reviewed by David Capwell for CASSANDRA-19336
Co-authored-by: Andrés de la Peña <a.penya.garcia@gmail.com>
Co-authored-by: David Capwell <dcapwell@apache.org>
When a node is decommissioned, it triggers data transfer to other nodes.
During this transfer process, receiving nodes temporarily hold token ranges in a pending state.
However, the current cleanup process doesn't account for these pending ranges when calculating token ownership,
leading to inadvertent cleanup of data already stored in SSTables.
To address this issue, this patch introduces two changes.
Firstly, it backports CASSANDRA-16418, introducing a preventive check in `StorageService#forceKeyspaceCleanup`.
This check disallows the initiation of cleanup when a node contains any pending ranges for the requested keyspace.
Secondly, it reintroduces a similar condition to test for the existence of pending ranges in `CompactionManager#performCleanup`.
This ensures the safety of this API as well.
Patch by Szymon Miezal; reviewed by Brandon Williams, Jacek Lewandowski for CASSANDRA-18824
Co-authored-by: Szymon Miezal <szymon.miezal@datastax.com>
Co-authored-by: Jacek Lewandowski <lewandowski.jacek@gmail.com>
This commit fixes Gossiper::hasMajorVersion3Nodes so that it does not return true when all hosts have a known version, no hosts are on a version earlier than 4.0, and there is a 4.x minor version or patch version upgrade in progress. Additionally, this commit improves the clarity of Gossiper::hasMajorVersion3Nodes's name to indicate that it will return true when the cluster has 3.x nodes or if the cluster state is unknown, matching the description in the in-line comment.
patch by Isaac Reath; reviewed by Paulo Motta and Stefan Miklosovic for CASSANDRA-18999
Closes#2967
Fixed a bug causing the `OpenDataException` being thrown
when executing the `repair_admin summarize-pending` command.
This patch addresses the problem by including a missing composite
in `PendingStats.toComposite`, ensuring proper data conversion.
patch by Szymon Miezal; reviewed by Stefan Miklosovic and Brandon Williams for CASSANDRA-19014
This is the follow-up commit of CASSANDRA-18935 where we set RPC_READY to false
when transports were shut down in runtime. The problem is that the current logic in StorageProxy.findSuitableEndpoint method,
used for the selection of a leader for counter mutations, is filtering out all endpoints which do not have RPC_READY
set to true. Hence, if there is a deployment of a coordinator node (not joining a ring) and storage nodes which
have transports turned off (e.g. for security reasons), then a coordinator node will select no endpoint as a
counter mutation leader which renders counter mutations impossible.
This change just reverts the original fix which was setting RPC_READY to false when transports were shut down in runtime
(e.g. by nodetool disablebinary).
For trunk (to be 5.1 at time of writing this), there is already TCM in place and the correct fix yet to be implemented
is to decouple from checking RCP_READY state and base it e.g. on JOINED state from TCM perspective. Please follow
CASSANDRA-19103 where this will be addressed.
patch by Stefan Miklosovic; reviewed by Brandon Williams for CASSANDRA-18935
HintsMaker was creating hint files which ttl'ed after 10 days which started to fail the tests.
By setting it explicitly to Integer.MAX_VALUE, newly generated hint files will expire in cca 68 years.
Branches 3.0 to trunk introduce the fix to HintsMaker. Branches 5.0 and trunk contain regenerated hint files as well.
Patches for 3.0, 3.11, 4.0 and 4.1 are just adding HintsMaker class to generate hint files.
Patches for 5.0 and trunk are adding tests for commit log upgrading and hints which were generated by HintsMaker.
Patches for 5.0 and trunk are also adding version 1 in HintsDescriptor back as it was removed in CASSANDRA-18314 by mistake.
patch by Stefan Miklosovic; reviewed by Michael Semb Wever for CASSANDRA-19002
The fix makes the DescribeStatement to wait for the in-progress schema transformations to finish before returning the first page. This way, the metadata and schema version encoded in the result set metadata are guaranteed to be consistent.
Patch by Jacek Lewandowski; reviewed by Benjamin Lerer, Ekaterina Dimitrova for CASSANDRA-18921
In this commit the `org.apache.cassandra.io.sstable.CQLSSTableWriter#build` method synchronizes on the
`Schema.instance` object (instead of the `CQLSSTableWriter.class`) to prevent concurrent schema operations
to fail when the offline tools also updates the schema.
For example, a table creation operation, which modifies the keyspace tables metadata, might end up
missing the update when a concurrent call to the `CQLSSTableWriter#build` method is accessing the
singleton Schema instance.
Patch by Francisco Guerrero, reviewed by Yifan Cai, Maxwell Guo, Alex Petrov for CASSANDRA-18317.
This patch also removes compress-lzf library (leftover from CASSANDRA-12229)
as well as it makes byte-buddy dependencies to be test scoped.
patch by Stefan Miklosovic; reviewed by Michael Semb Wever for CASSANDRA-18877
There are a couple of fixes in this patch. As explained on the ticket, some of the flaky failures are a race of two events that cause the termination of the connection. First is a legitimate close as a result of expected failures. In this case, the server sends the error message and closes the connection. The test expects to receive that message. However, the test was sending more messages that couldn't be received because the server already closed the connection, and they were bounced by the OS, causing immediate connection shutdown on the client side, even before it could receive the error message.
The fix is to stop sending the messages after sending the message, which is expected to cause a failure. Some other accompanying modifications include using Awaitility to wait for specific events and consider the configured maximum number of consecutive failures.
Also added some more logging to help investigate failures in the future.
Patch by Jacek Lewandowski; reviewed by Sam Tunnicliffe for CASSANDRA-16949
- Don't create uber jar for microbenchmarks
- Add async profiler to jmh tests
- Benchmark classes names validation
- Add jmh.args property to make it possible passing extra args to JMH
- Add missing test/anttasks to idea configuration
Patch by Jacek Lewandowski; reviewed by Branimir Lambov, Maxim Muzafarov, Stefan Miklosovic for CASSANDRA-18871
patch by Aleksey Yeschenko; reviewed by Blake Eggleston, Sam Tunnicliffe, and Caleb Rackliffe for CASSANDRA-16905
Co-authored by Aleksey Yeschenko (aleksey@apache.org)
Co-authored by Josh McKenzie (jmckenzie@apache.org)
patch by Vincent White; reviewed by Stefan Miklosovic and Branimir Lambov for CASSANDRA-14204
Co-authored-by: Stefan Miklosovic <smiklosovic@apache.org>
patch by Stefan Miklosovic; reviewed by Jacek Lewandowski and Brandon Williams for CASSANDRA-16555
Co-authored-by: Jacek Lewandowski <lewandowski.jacek@gmail.com>
Co-authored-by: Paul Rütter <paul@blueconic.com>
Add a comment about the use of disabled TLSv1.1 with JDK 8 and higher to negotiatedProtocolMustBeAcceptedProtocolTest
Patch by Dan Jatnieks; reviewed by Ekaterina Dimitrova and Brandon Williams for CASSANDRA-18540
Patch by Ningzi Zhan and Tomasz Lasica; reviewed by brandonwilliams,
edimitrova and Maxwell Guo for CASSANDRA-17818
Co-Authored-By: Tomek Lasica <tomasz.lasica@datastax.com>
If an sstable is corrupted in a nasty way, we may read invalid cell sizes and try to read much more data for a row than we should. In rare scenarios this can lead even to OOMs.
This simple fix adds tracking and limiting the amount of data that is read per row. Row has its size stored in preamble which can be used as a limit. If the deserialization code tries to read more than that, it will simply fail with EOF which will prevent more serious problems later.
Patch by Jacek Lewandowski; reviewed by Berenguer Blasi and Maxwell Guo for CASSANDRA-18513
# Conflicts:
# src/java/org/apache/cassandra/db/rows/UnfilteredSerializer.java
Adds separate WARN level messages to the check itself in Directories.
patch by Henrik Ingo; reviewed by Mick Semb Wever, Brad Schoening, Claude Warren, Kan Maung, maxwellguo for CASSANDRA-18260
Dmitry Konstantinov