Technologies/Trino/trino.execution.time.p95
TrinoTrinoMetric

trino.execution.time.p95

95th percentile execution time
Dimensions:None
Available on:Native (1)
Interface Metrics (1)
Native
95th percentile query execution time in the last five minutes
Dimensions:None
Sources
trino.execution.QueryManager.ExecutionTime.FiveMinute.P95trino.io

Technical Annotations (57)

Configuration Parameters (23)
retry-policyrecommended: QUERY
Use QUERY for small queries to avoid task-level retry overhead
query.execution-policyrecommended: phased
Default policy maximizes utilization and minimizes wall time
optimizer.join-reordering-strategyrecommended: AUTOMATIC
Enables automatic join reordering based on cost estimates
join_distribution_typerecommended: BROADCAST
Forces broadcast joins for small dimension tables to eliminate shuffling
join_reordering_strategyrecommended: AUTOMATIC
Session-level setting for join reordering optimization
hive.orc.use-column-namesrecommended: true
Enables column name-based ORC reading for better column pruning
parquet.optimized-reader.enabledrecommended: true
Enables optimized Parquet reader for better columnar access
delta.parquet.optimized-reader.enabledrecommended: true
Enables optimized Parquet reader for Delta Lake tables
hive.parquet.optimized-reader.enabledrecommended: true
Session-level Parquet optimization
hive.parquet.predicate-pushdown.enabledrecommended: true
Pushes predicates to Parquet reader for early filtering
enable-dynamic-filteringrecommended: true
Enables dynamic filtering for partition pruning (Trino 330+)
enable_dynamic_filteringrecommended: true
Session-level dynamic filtering configuration
dynamic_filtering_max_per_driver_row_countrecommended: 2000
Maximum rows per driver for dynamic filtering
dynamic_filtering_max_per_driver_sizerecommended: 1MB
Maximum size per driver for dynamic filtering
exchange.http-client.max-content-lengthrecommended: 4GB
Key setting - default too low for large data transfers between workers
query.remote-task.max-error-durationrecommended: 10m
Allows workers more time before declaring transport errors fatal
query.remote-task.max-request-sizerecommended: 10GB
Maximum size for requests between coordinator and workers
query.max-memory-per-noderecommended: 6GB
Per-node memory allocation for query processing
exchange.http-client.request-timeoutrecommended: 30s to 60s
Maximum time for individual HTTP requests between workers
exchange.http-client.idle-timeoutrecommended: 60s to 2m
Keep-alive timeout for idle HTTP connections
exchange.http-client.max-connections-per-serverrecommended: 1000
Connection pool size per worker node
query.client.timeoutrecommended: 60m
Overall client timeout for query execution
query.max-execution-time
Max query execution time in config.properties; increase if legitimate queries are timing out
Error Signatures (3)
Encountered too many errors talking to a worker nodelog pattern
PAGE_TRANSPORT_TIMEOUTerror code
QUERY_CANCELEDerror code
CLI Commands (4)
EXPLAIN ANALYZE <query>diagnostic
ANALYZE TABLEdiagnostic
SELECT query_id, user, query, create_time, end_time, (end_time - create_time) AS duration_seconds FROM trino_events.trino_queries WHERE end_time IS NOT NULL ORDER BY duration_seconds DESC LIMIT 10;diagnostic
SELECT user, AVG(end_time - create_time) AS avg_duration FROM trino_events.trino_queries WHERE end_time IS NOT NULL AND create_time >= now() - interval '7' day GROUP BY user ORDER BY avg_duration DESC LIMIT 10;diagnostic
Technical References (27)
S3 prefixcomponentcardinalityconceptS3 StandardcomponentS3 Standard-IAcomponentRemoteSourcecomponentLocalExchangecomponentFragmentconcepttrino_events.trino_queriescomponentcreate_timecomponentend_timecomponentcost-based optimizercomponentbroadcast joinsconceptORCcomponentParquetcomponentresource group managementconceptdynamic filteringconceptpartition pruningconceptbucketingconceptbloom filterconceptexchange.http-clientcomponentcoordinatorcomponentworkercomponentfault-tolerant executionconceptbatch operationsconceptFINISHINGcomponentBLOCKEDcomponenttrino-python-clientcomponent
Related Insights (19)
TASK retry policy increases latency for short-running querieswarning
▸
Undersized cluster causes slow query performancewarning
▸
Slow query performance from large unpartitioned data scanswarning
▸
Query performance degradation from stale data accumulationinfo
▸
High fragment blocked time indicates data transfer bottleneckwarning
▸
Query phase timing reveals specific performance bottleneck typesinfo
▸
Long-running queries hog cluster resourceswarning
▸
All-at-once execution policy increases queue times and query wall timeinfo
▸
Stale table statistics cause suboptimal join ordering and query failureswarning
▸
Wide table scans without columnar optimization cause I/O bottlenecksinfo
▸
Query queuing and inconsistent response times from concurrent session contentionwarning
▸
Poor partitioning schemes create massive I/O overhead for analytical querieswarning
▸
ETL workloads suffer I/O bottlenecks and uneven worker resource utilizationwarning
▸
PAGE_TRANSPORT_TIMEOUT causes query failures after 1 minutecritical
▸
High volume of short queries on fault-tolerant cluster causes latencywarning
▸
Long-running queries cause resource contentionwarning
▸
All-at-once execution policy causes increased queue timeswarning
▸
Queries canceled when execution time exceeds configured limitwarning
▸
Query hangs in FINISHING state for 5 minutes after fault-tolerant retrywarning
▸