In #51612 we promoted log messages regarding unexpected network disconnects from TRACE to DEBUG. Some users have indicated that they'd like even more visibility on these by default, at least for the ones that are definite indications of something going wrong at the network level, which IMO is at least Connection reset and Connection timed out on Linux. Normally we shouldn't be seeing either of these so I'm opening this issue to propose moving them all the way up to WARN.
I'm not decided on whether to push for this change in 7.x or to wait for 8.0. The trouble with doing it in 7.x is that transport clients are still in use in these versions and clients may have less reliable connectivity to Elasticsearch than we expect for node-to-node connections, so some of the logs we emit might not be actionable for the server administrator.
We might need a sync discussion on this topic eventually but for now I'd prefer to gather async thoughts here first.
In #51612 we promoted log messages regarding unexpected network disconnects from
TRACEtoDEBUG. Some users have indicated that they'd like even more visibility on these by default, at least for the ones that are definite indications of something going wrong at the network level, which IMO is at leastConnection resetandConnection timed outon Linux. Normally we shouldn't be seeing either of these so I'm opening this issue to propose moving them all the way up toWARN.I'm not decided on whether to push for this change in 7.x or to wait for 8.0. The trouble with doing it in 7.x is that transport clients are still in use in these versions and clients may have less reliable connectivity to Elasticsearch than we expect for node-to-node connections, so some of the logs we emit might not be actionable for the server administrator.
We might need a sync discussion on this topic eventually but for now I'd prefer to gather async thoughts here first.