Libelektra: High-Level API: Test Seems to Stall Sometimes

[sanssecours]

Problem Description

It looks like the tests testkdb_highlevel stalls sometimes on the Jenkins build server, since it might fail with exit code 143. As far as I can tell from the Jenkins documentation, exit code 143 means the process was terminated with signal SIGTERM (143-128=15), which usually happens if ctest kills a test that hit the timeout limit.

System Information

• Elektra Version: master

Build Server Logs

• master
• PR #2545

[kodebach]

I don't think it stalls, testkdb_highlevel is just inherently a very slow test. Every elektraSet* call (of which a good test has many) includes a call to kdbSet and kdbGet and is therefore a bit slow. In combination with valgrind and high server load this results in pretty long test durations.

For example in the PR #2545 build just before the one you linked, it took nearly 4 minutes to complete the test in BUILD_FULL "optimizations off"-mode.

[sanssecours]

Thank you for the quick update. If the test really takes that long to execute, then it might make sense to just increase the test timeout?

[kodebach]

then it might make sense to just increase the test timeout?

Seems like the easiest solution, yes.

[sanssecours]

According to this Stackflow article and the source code of CMake the default CTest timeout should be 1500s (25 min). Do we limit this test timeout somewhere in the CMake code or does testkdb_highlevel really take that long to execute?

[kodebach]

I highly doubt it takes that long...

Also I looked at some of the problematic Jenkins jobs, and they all took a lot less than 25 min before they stopped. Here the build step in question aborted after only 5m 9s...

So the SIGTERM must come from something other than a timeout.

[markus2330]

Thank you for looking into this and fixing the problem!