Merge #138834 #140960 #140995

138834: kvserver,storeliveness: introduce follower quiescence r=arulajmani,iskettaneh a=miraradeva

This commit introduces the ability for followers on inactive ranges to quiesce and stop ticking in Raft. This will allow the CPU utilization of the cluster to decrease by a factor of 2/3 (assuming RF3).

Unlike epoch lease quiescence, which also quiesces the leader under certain conditions, store liveness quiescence is much simpler. The conditions to quiesce/unquiesce are:

- Quiesce a follower if it hasn't received Raft messages in a certain number of ticks and the follower's store provides store liveness support for the leader's store.

- Unquiesce a follower if it hears from store liveness that the follower's store has withdrawn support from the leader's store.

Fixes: #133885

Release note: None

140960: sql: deflake TestDropDatabaseDeleteData r=iskettaneh a=iskettaneh

This commit deflakes the test by blocking the full reconciliation from starting before we drop the database. This is used to avoid a race condition where the full reconciliation starts after we drop the database, it will ignore the dropped database. Then, there is a race between the SQLWatcher and the GC job where the SQLWatcher might write a zone config of table1 before table2, while the GC queue might not know that this range needs to be split because it has different span config.

Release note: None

Fixes: #138185

140995: sql: Fix trigger dependency cleanup in legacy schema changer r=spilchen a=spilchen

In the legacy schema changer for DROP TABLE, there was special logic to clean up dependencies created by triggers. However, a bug in this logic caused the dependent table’s descriptor to remain unchanged, leaving behind a dangling reference. This commit ensures that the dependent table’s descriptor is correctly updated.

Epic: none
informs https://github.com/cockroachlabs/support/issues/3174
Release note (bug fix): Fixed a bug where dropping a table with a trigger using the legacy schema changer could leave an orphaned reference in the descriptor. This occurred when two tables were dependent on each other via a trigger, and the table containing the trigger was dropped.

Co-authored-by: Mira Radeva <[email protected]>
Co-authored-by: Ibrahim Kettaneh <[email protected]>
Co-authored-by: Matt Spilchen <[email protected]>

Author: 4 people

docs/generated/metrics/metrics.html

@@ -620,6 +620,7 @@
 <tr><td>STORAGE</td><td>rebalancing.writebytespersecond</td><td>Number of bytes written recently per second, considering the last 30 minutes.</td><td>Bytes/Sec</td><td>GAUGE</td><td>BYTES</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>STORAGE</td><td>rebalancing.writespersecond</td><td>Number of keys written (i.e. applied by raft) per second to the store, considering the last 30 minutes.</td><td>Keys/Sec</td><td>GAUGE</td><td>COUNT</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>STORAGE</td><td>replicas</td><td>Number of replicas</td><td>Replicas</td><td>GAUGE</td><td>COUNT</td><td>AVG</td><td>NONE</td></tr>
+<tr><td>STORAGE</td><td>replicas.asleep</td><td>Number of asleep replicas. Similarly to quiesced replicas, asleep replicas do not tick in Raft.</td><td>Replicas</td><td>GAUGE</td><td>COUNT</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>STORAGE</td><td>replicas.leaders</td><td>Number of raft leaders</td><td>Raft Leaders</td><td>GAUGE</td><td>COUNT</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>STORAGE</td><td>replicas.leaders_invalid_lease</td><td>Number of replicas that are Raft leaders whose lease is invalid</td><td>Replicas</td><td>GAUGE</td><td>COUNT</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>STORAGE</td><td>replicas.leaders_not_fortified</td><td>Number of replicas that are not fortified Raft leaders</td><td>Replicas</td><td>GAUGE</td><td>COUNT</td><td>AVG</td><td>NONE</td></tr>
@@ -844,6 +845,7 @@
 <tr><td>STORAGE</td><td>storage.write-amplification</td><td>Running measure of write-amplification.<br/><br/>Write amplification is measured as the ratio of bytes written to disk relative to the logical<br/>bytes present in sstables, over the life of a store. This metric is a running average<br/>of the write amplification as tracked by Pebble.</td><td>Ratio of bytes written to logical bytes</td><td>GAUGE</td><td>COUNT</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>STORAGE</td><td>storage.write-stall-nanos</td><td>Total write stall duration in nanos</td><td>Nanoseconds</td><td>GAUGE</td><td>NANOSECONDS</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>STORAGE</td><td>storage.write-stalls</td><td>Number of instances of intentional write stalls to backpressure incoming writes</td><td>Events</td><td>GAUGE</td><td>COUNT</td><td>AVG</td><td>NONE</td></tr>
+<tr><td>STORAGE</td><td>storeliveness.callbacks.processing_duration</td><td>Duration of support withdrawal callback processing</td><td>Duration</td><td>HISTOGRAM</td><td>NANOSECONDS</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>STORAGE</td><td>storeliveness.heartbeat.failures</td><td>Number of Store Liveness heartbeats that failed to be sent out by the Store Liveness Support Manager</td><td>Heartbeats</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
 <tr><td>STORAGE</td><td>storeliveness.heartbeat.successes</td><td>Number of Store Liveness heartbeats sent out by the Store Liveness Support Manager</td><td>Heartbeats</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
 <tr><td>STORAGE</td><td>storeliveness.message_handle.failures</td><td>Number of incoming Store Liveness messages that failed to be handled by the Store Liveness Support Manager</td><td>Messages</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>

pkg/ccl/logictestccl/testdata/logic_test/triggers

@@ -4030,4 +4030,54 @@ drop_trigger
 statement ok
 CREATE OR REPLACE FUNCTION f() RETURNS TRIGGER LANGUAGE PLpgSQL AS $$ BEGIN RETURN NEW; END $$;
 
+# ==============================================================================
+# Test that descriptor backreferences are cleaned up
+# ==============================================================================
+
+subtest ensure_back_ref_cleanup
+
+statement ok
+create table listings_balance (c1 int);
+
+# Create a function that references listings_balance
+statement ok
+create or replace function update_listing_balance()
+returns TRIGGER language PLpgSQL AS
+$$
+BEGIN
+  INSERT INTO listings_balance VALUES (1);
+  RETURN NEW;
+END
+$$;
+
+statement ok
+CREATE TABLE transaction_entries (c1 int);
+
+# Create a trigger that references listings_balance through the trigger function.
+statement ok
+create trigger tr AFTER INSERT OR UPDATE ON transaction_entries FOR EACH ROW execute function update_listing_balance();
+
+# Force the legacy schema change for the drop table. We will save the original
+# DSC setting so that it be reset properly.
+let $use_decl_sc
+SHOW use_declarative_schema_changer
+
+statement ok
+SET use_declarative_schema_changer = off;
+
+# This should cleanup the dependency in listings_balance.
+statement ok
+DROP TABLE transaction_entries;
+
+# Reinstate original DSC setting
+statement ok
+SET use_declarative_schema_changer = $use_decl_sc;
+
+statement ok
+DROP FUNCTION update_listing_balance ;
+
+# Sanity to verify the dependency no longer exists.
+statement ok
+drop table listings_balance cascade;
+
 subtest end

pkg/kv/kvserver/BUILD.bazel

@@ -78,6 +78,7 @@ go_library(
         "replica_split_load.go",
         "replica_sst_snapshot_storage.go",
         "replica_store_liveness.go",
+        "replica_store_liveness_sleep.go",
         "replica_tscache.go",
         "replica_write.go",
         "replicate_queue.go",