You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Deprecated Dashboard: the blocked-process and deadlock collectors shred the whole XE ring buffer on every run #4213
#4200 found that the blocked-process and deadlock collectors convert and shred their whole Extended Events ring buffer on every run. Almost every run finds nothing new. PR #4212 fixes this for Darling and Lite. Both now read the session target's execution_count first, and skip the shred when it has not changed since the last run.
The deprecated Dashboard's SQL Agent collectors have the same pattern, and #4212 leaves them unchanged:
install/22_collect_blocked_processes.sql
install/24_collect_deadlock_xml.sql
Each one converts target_data to XML and shreds ring_buffer.nodes(...) on every run, on the monitored server.
Store the target's last execution_count, and skip the conversion and the shred when the new count matches it.
Read the whole buffer on the first run, when the count goes down (the session restarted), or when no count is available.
The rows stored must stay the same.
Decision needed
The Dashboard is deprecated, so someone has to decide whether it still gets performance fixes. The fix changes the installed collector procedures and needs a place to keep the last count, so it also needs install and upgrade testing.
Activity
erikdarlingdata commentedon Sep 25, 2026
Queued at tier 4, at Erik's word (2026-09-25): the deprecated Dashboard gets this fix too. It will be worked after this wave's PRs land.
Add two upgrade-path pins and fix an undeclared @shred_needed in the …
Add source-text pins for the dedupe predicate and dynamic-SQL DECLARE…
Gate the deprecated Dashboard's XE ring-buffer shred on execution_cou…
erikdarlingdata commentedon Sep 26, 2026
Fixed by #4371, merged to dev (dfe9288).