Splice the pending CHANGELOG entries merged through 2026-09-28T17:40:00Z UTC

[erikdarlingdata]

Moves the CHANGELOG entries carried in merged PR descriptions into [Unreleased]. Cut: PRs merged to dev after 2026-09-26T17:37:33Z UTC (the #4440 cut) and at or before 2026-09-28T17:40:00Z UTC; the next splice starts after it.

What changes

One Python script (not committed, run outside the repo) listed every closed PR with base dev by last update, newest first, until it passed the window start, and kept the 97 PRs merged inside the window. It read each PR's ## CHANGELOG block and spliced it into [Unreleased] in blob space (the committed blob is LF-only).

Spliced: 76 PRs (each counted once, under its first section):

• Fixed: 36 (38 bullets) - [Retry the alert pass's database-state maintenance once on a chunk-drop deadlock #4612, Lite and the Darling Viewer: every clipboard write retries instead of crashing when the clipboard is busy (#4594) #4600, Plan viewer: guard clipboard writes, and hand back the captured query on a truncated single-statement copy #4593, get_ag_health fills its fleet-wide answer to the 32 KB budget by size, not a fixed group count #4568, Plan analysis: the repro script matches PerformanceStudio's (#4564) #4567, Stop swallowing a cancelled request in 8 catch blocks #4562, Lite: a cancelled analysis request stops instead of reading to the end #4561, Plan analysis: parsing and analysis stop when the caller cancels, and the async path enforces the size limit (#4512) #4560, Plan analysis: statements inside a procedure, function or cursor sub-plan now get findings (#4514) #4557, Plan analysis: the non-SARGable check reads which side and which table a function or conversion is on (#4518, #4519) #4554, Store: re-measure a rollup's coverage floor only when its oldest chunk changes (#4539) #4553, Plan analysis: run the benefit scorer at every entry point and show each finding's benefit (#4546) #4552, Plan analysis: a deeply nested plan no longer crashes the process that parses it (#4512) #4551, Plan analysis: adaptive join note, truncated-text rule, UDF/serial dedup, invariant percentages (#4531, #4532, #4533, #4536) #4548, Plan analysis: operator self-time excludes the coordinator thread and looks through Compute Scalar and batch mode subtrees (#4513, #4515) #4547, Plan analysis: Join OR Clause rule requires a lookup that reads another input (#4521) #4544, Plan analysis: skip Row Estimate Mismatch on unexecuted operators, skip engine functions in the TVF rule (#4522, #4525) #4542, Store: give each heavy table its own compression hour so the nightly pass stops timing out collectors (#4510) #4541, Plan analysis: mask comments and string literals before the text rules (#4524) #4540, Darling keeps retrying a slow managed PostgreSQL start instead of never collecting (#4508) #4538, Darling keeps retrying the store at start instead of never collecting after the retry budget (#4508) #4509, After a job-history identity reset, the collector re-reads once instead of on every run (#4487) #4496, A missing hour in the collection-health rollup no longer sends every fleet health read to a 7-day raw scan (#4477) #4494, Lite's perfmon chart and get_perfmon_trend set aside a one-sample Wait Statistics spike (#4476) #4492, Perfmon trend charts and get_perfmon_trend set aside a one-sample Wait Statistics spike instead of plotting it (#4476) #4490, Lite's Job History shows that it is loading and says when it shows only the newest rows (#4478) #4488, query_store's normal hourly source cadence no longer reads as "produced then stopped" (#4473) #4483, The Viewer counts Availability Groups by identity, shows Job History loading, and says "queries" (#4475, #4478) #4481, The plan parser reads IF-condition query plans and MULTIPLE PLAN statement hashes (#4468) #4470, The upgrade script waits for the service's process to exit before refusing (#4466) #4467, A hand-edited or interrupted settings-file migration heals on stores that expose the managed PostgreSQL on the network #4465, Stores that expose the managed PostgreSQL on the network finish the settings-file migration #4464, The chunk-interval adjustment's 1,000-chunk limit applies to each table's own chunks (#4457) #4458, Severe-error rows read the database id from the event's action and drop what sp_HealthParser drops (#4452) #4455, A carried maintenance_work_mem is capped for PostgreSQL 17 and earlier when the managed settings move (#4336) #4444, After a long outage across an upgrade, the seam repair also refreshes the successor daily rollup (#4300) #4439]
• Changed: 28 - [Bound the viewer/mcp roles' on-disk temp-file spill (#4605) #4610, Plan viewer: minimap resize, double-click zoom and coloured edges (#4580) #4603, Plan analysis: rule 38 reads the server's edition and MAXDOP at every entry point (#4530) #4601, Give the Plan Insights strip one card system instead of per-card tints #4599, Port PerformanceStudio's wait-row layout: an Auto benefit column that never clips #4595, Plan viewer: a shared MetricFormatter for costs and durations #4592, Plan viewer: the Wait Stats header's tooltip matches PerformanceStudio #4591, Match PerformanceStudio's runtime summary card: title, order, memory-grant colors, spill flag #4590, Plan viewer: properties panel row model, filter box, copy menu, remembered width #4588, Roll the per-thread stats into one collapsed breakdown per section #4587, Plan viewer: color edges by the actual/estimated row ratio (#4579) #4586, Wait-category colours: a laid-out, contrast-checked palette #4583, Plan analysis: drop three rules PerformanceStudio removed, and show its CPU:Elapsed ratio (#4537) #4565, Bump the nuget-patch-and-minor group with 1 update #4563, Store rung V152: drop the Query Store rollups' unread group indexes #4506, Store: refresh a plan or text dimension's last_seen at most every 6 hours, not every hour #4502, Darling: name each session's application in the store's own log (managed-conf v17) #4501, Store rung V151: collect each Availability Group's group_id and count groups by it (#4475) #4495, The Viewer's Job History reads each server's newest runs through an index (#4477) #4493, Store rung V150: a per-collector watermark index and a per-server job-history index (#4469, #4477) #4489, Every Darling store connection names its surface in application_name (#4479) #4486, The Viewer's Overview shares one fleet-health read across its server cards, and the status bar stops re-measuring the store on every refresh (#4477) #4482, The startup watermark read reads only the last two days of collection_log, bound as a naive timestamp (#4469) #4480, get_ag_health caps its fleet-wide answer, most severe first, and says when it cut (#4471) #4474, The Query Store liveness touch updates rows in place: drop the last_seen indexes, fillfactor 90 (#4250) #4472, Scheduler issues read the utilization fields the event carries, flagged the way sp_HealthParser flags them (#4452) #4456, Raise the viewer and mcp roles' default statement_timeout from 15 s to 60 s (#4442) #4447, Add a v16 managed-conf marker for a 15-minute checkpoint interval (#4246) #4426]
• Added: 12 (13 bullets) - [Plan viewer: add a Server Context card to the Plan Insights strip (#4597) #4614, Plan viewer: add a Parameters card to the Plan Insights strip #4604, Plan viewer: add the minimap (#4580 part 1) #4596, Plan analysis: PerformanceStudio's per-rule configuration, rule numbers, rule 38 and the legacy marker (#4535, #4566) #4585, Plan viewer: a finding's header links to the operator it came from (#4534) #4559, Plan analysis: report which operator each finding came from (#4534) #4556, Plan analysis: wait stats become findings, and external and preemptive waits get a real benefit (#4516, #4517) #4555, Plan analysis: flag an expensive operator no other rule caught (#4527) #4550, Plan analysis: dynamic-cursor and missing-LOCAL cursor rules (#4529) #4549, Plan analysis: Bare Scan rule and Serial Plan benefit formula (#4526, #4528) #4545, Plan analysis: mark which warnings are SQL Server's and key duplicate index suggestions on the database (#4520, #4523) #4543, Add get_read_latency: p50/p95/p99 durations per web and MCP read, and name the web and MCP store connections (#4442) #4454]

Each entry was inserted at the top of its section's list, newest PR number first, matching the file's own order. Each REF: definition was appended to the trailing definitions block in ascending order, skipped when the label already existed.

Multi-entry PRs

• Plan analysis: adaptive join note, truncated-text rule, UDF/serial dedup, invariant percentages (#4531, #4532, #4533, #4536) #4548 carries two groups: a Fixed entry and an Added entry (the truncated-query-text finding). Both are spliced, so Added gains 13 bullets: one from each of its 12 PRs, plus Plan analysis: adaptive join note, truncated-text rule, UDF/serial dedup, invariant percentages (#4531, #4532, #4533, #4536) #4548's. Plan analysis: adaptive join note, truncated-text rule, UDF/serial dedup, invariant percentages (#4531, #4532, #4533, #4536) #4548 is counted once, under Fixed.
• The Viewer counts Availability Groups by identity, shows Job History loading, and says "queries" (#4475, #4478) #4481's single Fixed entry holds three bullets. They are kept together, in their order.

Mechanical fixes

• Darling keeps retrying a slow managed PostgreSQL start instead of never collecting (#4508) #4538 carried a second REF, [#4509], that Darling keeps retrying the store at start instead of never collecting after the retry budget (#4508) #4509's own block also carries with the same URL. It is defined once.
• Plan analysis: adaptive join note, truncated-text rule, UDF/serial dedup, invariant percentages (#4531, #4532, #4533, #4536) #4548's two groups each carried [#4548]. It is defined once.
• Plan analysis: rule 38 reads the server's edition and MAXDOP at every entry point (#4530) #4601's block ends with a scrubcheck: clean. line. It is not an entry and is not spliced.
• Every REF label matched its URL's number, and every URL is a /pull/ link to that PR, so no URL needed a fix.

IMPORTANT notes (not spliced)

Five PRs carry an IMPORTANT: upgrade note after their REF. [Unreleased] has no ### Important section, and the release checklist writes that section at the cut, so these are listed here for the cut and are not in this diff:

• Store rung V150: a per-collector watermark index and a per-server job-history index (#4469, #4477) #4489: Store rung V150 builds two indexes on first start after upgrade; collection_log and job_history writes wait for the build (about a second per 2.4 million uncompressed rows on the test rig).
• Store rung V151: collect each Availability Group's group_id and count groups by it (#4475) #4495: Store rung V151 adds a group_id column to ag_replica_states and ag_database_replica_states (Lite: schema 65). Rows collected before the upgrade have it empty.
• Store rung V152: drop the Query Store rollups' unread group indexes #4506: Store rung V152 drops unused indexes on six Query Store rollups (no data changes; rows read the same before and after). During a one-time upgrade start that lands on a running rollup refresh, Query Store trend reads can wait up to ~4.5 minutes, and collection starts after the upgrade finishes.
• Store: give each heavy table its own compression hour so the nightly pass stops timing out collectors (#4510) #4541: On the first start after upgrading, the store retunes the compression delay of its nine heaviest tables in place. Those tables keep up to 9 more hours of data uncompressed, so the store's disk use rises slightly. Nothing needs to be done.
• Darling: name each session's application in the store's own log (managed-conf v17) #4501: An existing managed store's postgresql.conf gains a new block on its next service-owned start (or darling-managed.conf's rendered body gains the setting, on a store that has moved to that file). No setting already in force through ALTER SYSTEM is touched.

Skips (21 PRs - no user-visible entry)

Each of these says SECTION: None or opens its CHANGELOG block with a free-text None: line (test-only, CI-only, analyzer or warning cleanup, or an effect that has no user-visible surface yet): #4440, #4441, #4443, #4445, #4446, #4448, #4449, #4451, #4453, #4460, #4461, #4462, #4463, #4491, #4485, #4500, #4504, #4558, #4581, #4584, #4607.

• Splice the pending CHANGELOG entries merged through 2026-09-26T17:37:33Z UTC #4440 is the previous splice PR itself.
• Record web and composed-panel read durations into hourly latency histograms in the store (#4442) #4451 and Record every MCP tool call's duration into the read-latency histograms (#4442) #4453 record read latencies that Add get_read_latency: p50/p95/p99 durations per web and MCP read, and name the web and MCP store connections (#4442) #4454 (spliced, Added) surfaces.
• Plan analysis: the store readers fill ServerMetadata's cost threshold, max server memory and database settings (#4597) #4607 feeds the Server Context card that Plan viewer: add a Server Context card to the Plan Insights strip (#4597) #4614 (spliced, Added) shows.
• Plan viewer: show each warning's actionable fix, and take severity colours from one helper #4584's fix line renders only for a warning that carries an actionable fix, and no rule sets one yet.

No PR number in this window was already cited in [Unreleased] or already defined in the file.

Self-check

• The diff is CHANGELOG.md only: 155 insertions, 0 deletions (79 bullets and 76 definitions).
• The staged blob is byte-identical to the script's output, and it stays LF-only (0 CR bytes). The working tree was re-checked-out from the index, so it is CRLF as .gitattributes says.
• Every [#N] cited in [Unreleased] has exactly one definition, the file has no duplicate labels, and every definition's label matches its URL's number.
• python tools/changelog/changelog_archive.py verify: exit 0. Its INFO line lists 5 unreferenced definitions (Lite Overview baseline bands use the server's actual hour under a custom range (#4320) #4325, Size the Linux compose store's workers and work_mem from the product's own formula (#4322) #4400, Hold the raw purge on any hole in the frozen legacy's span, and fill the successor down to raw's floor (#4301) #4401, Charts should show visible flat line at zero instead of empty #85, Deadlock pipeline latency: too many sequential hops before data reaches charts #86); none of them comes from this splice.

Test plan

• Built Darling/Darling.Tests/Darling.Tests.csproj (Debug): 0 warnings, 0 errors.
• Darling.Tests.exe -class Darling.Tests.ChangelogIndexAndArchiveTests: Total 5, Failed 0, Skipped 0.

CHANGELOG

SECTION: None — this PR is the CHANGELOG change itself.