SOLR-16954: Fix test failure in UPDATE circuit breaker

[janhoy]

https://issues.apache.org/jira/browse/SOLR-16954

The tests ran fine both on localhost and github actions, but on Jenkins this one tripped during indexing content in indexDocs(). Must be some difference in memory handling, perhaps how many tests are ran in parallel in Jeknins vs GitHub actions/Crave, or the -Xmx setting??

Output from failing test:

Seen JVM heap memory usage 383.848.016 and allocated threshold 383.778.816

383Mb threshold suggests that the VM is ran with 512Mb RAM, which makes sense.

[janhoy]

I guess this failing test also begs the question whether the memory breaker is too trigger happy, tripping for a spike during a ms memory pressure. If we had an average memory pressure over N seconds that would be so much better. Does not dropwizard provide such, or would we need to roll our own?

[janhoy]

I think the intention of including the update breaker in the config was to exercise the initialization. So another approach could be to keep it but increase the threshold to 99% or something...

[cpoerschke]

... keep it but increase the threshold to 99% or something...

I like that idea.

And if it still causes regular issues at 99% or so threshold then removal is still an option. Or then another option could be to sub-class the update handler, configure the sub-classed update handler and have it ignore the circuit breaker when some magic field is part of the test setup update requests.