Merge #139735

139735: cli: improve debug zip transaction_contention_events query r=kyle-a-wong a=kyle-a-wong

Previously, this query often timed out with errors like "memory budget exceeded" and "query execution canceled due to statement timeout." This was happening because the query needed to deduplicate all the rows returned after performing two left joins on system.statement_statistics. This table is denormalized, meaning many rows are returned for every row joined.

To improve this query, CTEs are used to first create distinct (statement_fingerprint, query) and (transaction_fingerprint, queries) tables. Then, transaction_contention_events is joined onto those tables. This reduces the amount of deduplication required, improving performance.

Fixes: CRDB-45216
Epic: none
Release note (cli change): Improves the performance of the debug zip query that collects transaction_contention_events data, reducing the chances of "memory budget exceeded" or "query execution canceled due to statement timeout" errors.

Co-authored-by: Kyle Wong <[email protected]>

Author: craig[bot]

pkg/cli/zip_table_registry.go

@@ -550,26 +550,37 @@ var zipInternalTablesPerCluster = DebugZipTableRegistry{
 	},
 	"crdb_internal.transaction_contention_events": {
 		customQueryUnredacted: `
+with fingerprint_queries as (
+	SELECT distinct fingerprint_id, metadata->> 'query' as query  
+	FROM system.statement_statistics
+),
+transaction_fingerprints as (
+		SELECT distinct fingerprint_id, transaction_fingerprint_id
+		FROM system.statement_statistics
+), 
+transaction_queries as (
+		SELECT tf.transaction_fingerprint_id, array_agg(fq.query) as queries
+		FROM fingerprint_queries fq
+		JOIN transaction_fingerprints tf on tf.fingerprint_id = fq.fingerprint_id
+		GROUP BY tf.transaction_fingerprint_id
+)
 SELECT collection_ts,
        contention_duration,
        waiting_txn_id,
        waiting_txn_fingerprint_id,
        waiting_stmt_fingerprint_id,
-       s.metadata ->> 'query'                      AS waiting_stmt_query,
+       fq.query                      AS waiting_stmt_query,
        blocking_txn_id,
        blocking_txn_fingerprint_id,
-       array_agg(distinct ss.metadata ->> 'query') AS blocking_txn_queries_unordered,
+       tq.queries AS blocking_txn_queries_unordered,
        contending_pretty_key,
        index_name,
        table_name,
        database_name
 FROM crdb_internal.transaction_contention_events
-         LEFT JOIN system.statement_statistics AS s ON waiting_stmt_fingerprint_id = s.fingerprint_id
-         LEFT JOIN system.statement_statistics AS ss ON ss.transaction_fingerprint_id = blocking_txn_fingerprint_id
-WHERE ss.transaction_fingerprint_id != '\x0000000000000000' AND s.fingerprint_id != '\x0000000000000000'
-GROUP BY collection_ts, contention_duration, waiting_txn_id, waiting_txn_fingerprint_id, blocking_txn_id,
-         blocking_txn_fingerprint_id, waiting_stmt_fingerprint_id, contending_pretty_key, s.metadata ->> 'query',
-         index_name, table_name, database_name
+LEFT JOIN fingerprint_queries fq ON fq.fingerprint_id = waiting_stmt_fingerprint_id
+LEFT JOIN transaction_queries tq ON tq.transaction_fingerprint_id = blocking_txn_fingerprint_id
+WHERE fq.fingerprint_id != '\x0000000000000000' AND tq.transaction_fingerprint_id != '\x0000000000000000'
 `,
 		customQueryUnredactedFallback: `
 SELECT collection_ts,