In a recent support incident we noticed nodes with very large numbers of connections to s3. It was, however, difficult to determine how many connections the nodes were intended to have open during the restore operation, i.e. how many files intersected the restore span entries being processed. Logging per entry could be useful to note number of files that will be opened, and when the entry closes its files, though this could be verbose in large restores if on by default. Tracing events could be useful, or potentially more instrumentation in the cloud io packages.
In a recent support incident we noticed nodes with very large numbers of connections to s3. It was, however, difficult to determine how many connections the nodes were intended to have open during the restore operation, i.e. how many files intersected the restore span entries being processed. Logging per entry could be useful to note number of files that will be opened, and when the entry closes its files, though this could be verbose in large restores if on by default. Tracing events could be useful, or potentially more instrumentation in the cloud io packages.
Jira issue: CRDB-36256