Release Notes for MariaDB Enterprise ColumnStore 23.02.3

Can result in a hang or crash

• When the GROUP_CONCAT() function is used to produce relatively long strings, excessive memory can be used, which can cause PrimProc to crash with a segmentation fault. (MCOL-5385, MCOL-5429)

  • In previous releases, this issue can cause various types of messages to appear in the ColumnStore logs prior to the crash, such as:

    CAL0005: ThreadPool: Caught exception during execution:  boost::thread_resource_error: Resource temporarily unavailable [generic:11]
    CAL0000: TupleAggregateStep::doThreadedAggregate() caught std::bad_alloc         %%10%%
    

  • In previous releases, this issue can cause clients to see the following error message:

    ERROR 1815 (HY000): Internal error: InetStreamSocket::readToMagic: Remote is closed
    

  • Starting with this release, less memory should be used in this scenario.

• When multiple ColumnStore threads or processes try to access a managed shared memory segment for the Extent Map concurrently, a read/write conflict can cause the processes to crash with a segmentation fault due to a race condition. (MCOL-5487)

  • In previous releases, cpimport, PrimProc, and DMLProc were observed to crash in this scenario.

  • Starting with this release, the internal RWLock is upgraded to a write lock when a shared memory segment remap is initiated (such as by a long running cpimport job), thus avoiding this read/write conflict.