Currently with the addition of database ID to the sampled query telemetry log, we require access to descriptors every time we attempt to emit such a log. However, if these descriptors or the leasing mechanism are unavailable, the emission of this log will indefinitely block, greatly impacting the corresponding SQL Session.

In response to this issue, and importantly, the broader issue:
How can we ensure that SQL continues to perform normally even when the cluster may be partially unavailable?

and the issue in relation to SQL Observability:
How can we ensure that SQL Observability features allow SQL to continue to perform normally even when the cluster may be partially unavailable?

we would like to introduce a new session variable for a "troubleshooting" mode, where if enabled (disabled by default), queries do not get additional work attached to them.

Jira issue: CRDB-17655