Several years ago, all tests, except bigmem tests could be ran sequentially with just 650 MB of memory. There were issues with some buldbots (e.g. AIX) which small RAM, because excess memory consumption not always caused a quick test failure, but could cause swapping and failing tests at timeout 2 hours later. So all tests with high memory consumption were decorated withbigmemtest.

Now, some tests need more than 1 GB of memory, or even more than 2 GB. Not all buildbots have such much.

The main culprit is running too many threads. Every thread needs 35, 70 or even 90 MB, depending on test.

There are two possible solutions:

1. Decrease the number of threads. Some tests can perform their function with fewer threads, but this must be decided individually so as not to make the test useless.
2. Decorate the test withbigmemtest. This will skip it on most buildbots.

Linked PRs

• gh-133454: Mark tests with many threads that use much memory as bigmem #133456
• gh-133454: Reduce the number of threads in test_racing_getbuf_and_releasebuf #133458
• [3.14] gh-133454: Mark tests with many threads that use much memory as bigmem (GH-133456) #133663
• [3.13] gh-133454: Mark tests with many threads that use much memory as bigmem (GH-133456) #133664
• gh-133454: Mark test_queue tests with many threads as bigmem #134575
• [3.14] gh-133454: Reduce the number of threads in test_racing_getbuf_and_releasebuf (GH-133458) #134589