basebackup with high slru count prepares a single tenant which is then templated to multiple tenants, and those are being basebackupped. In this preparation phase in the test there shutdown of the NeonEnv.
The environment assumption is that we run the benches on our small runners and we do not overprovision the runners (for example, running two unrelated github jobs on the same instance). Is it plausible that the runner had really a pause of 10 seconds OR is it more likely that the previous benchmarks had left the system in such a state that it required 10 seconds for writing out the dirty bytes?
Example:
basebackup with high slru count prepares a single tenant which is then templated to multiple tenants, and those are being basebackupped. In this preparation phase in the test there shutdown of the
NeonEnv
.The environment assumption is that we run the benches on our small runners and we do not overprovision the runners (for example, running two unrelated github jobs on the same instance). Is it plausible that the runner had really a pause of 10 seconds OR is it more likely that the previous benchmarks had left the system in such a state that it required 10 seconds for writing out the dirty bytes?