From 653436460338b75aaacb3921a0fe5e1cbcaba9df Mon Sep 17 00:00:00 2001 From: Nate Coraor Date: Tue, 29 Oct 2024 14:14:24 -0500 Subject: [PATCH] Add query for memory efficiency of tools --- parts/22-query.sh | 79 +++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 79 insertions(+) diff --git a/parts/22-query.sh b/parts/22-query.sh index 2226b6e..e652bbd 100644 --- a/parts/22-query.sh +++ b/parts/22-query.sh @@ -5572,3 +5572,82 @@ query_tools-usage() { ##? [year] [--tools=] [--short_tool_id] [ $tool_id EOF } + +query_tool-memory-efficiency() { ##? [--newer-than=2592000] [--min-job-count=5] [--min-runtime=300] [--min-mem=8]: View the efficiency of recent jobs + meta <<-EOF + AUTHORS: natefoo + ADDED: 23 + EOF + handle_help "$@" <<-EOF + This is mainly useful for determining which long-running tools with large memory allocations have been + granted too large of an allocation. It requires that your jobs run in isolated Linux cgroups (as is + commonly configured for Slurm) that accurately report memory usage in memory.peak (cgroupsv2) or + memory.memsw.max_usage_in_bytes (cgroupsv1) and you have enabled the Galaxy cgroups job metrics plugin. + + $ gxadmin query tool-memory-efficiency + tool_id | run_count | total_runtime_hrs | avg_mem_used_gb | avg_mem_allocated_gb | avg_mem_percent + ---------------------+-----------+-------------------+-----------------+----------------------+----------------- + samtools_sort/2.0.5 | 232 | 48.65 | 10.00 | 10.00 | 100.00 + featurecounts/2.0.3 | 657 | 123.18 | 7.77 | 8.00 | 97.07 + rna_star/2.7.10b | 20 | 59.47 | 51.34 | 58.00 | 88.51 + kraken2/2.1.1 | 156 | 74.76 | 78.03 | 84.82 | 91.55 + bwa_mem2/2.2.1 | 635 | 399.82 | 53.24 | 57.82 | 90.01 + trim_galore/0.6.7 | 1217 | 358.12 | 10.74 | 18.50 | 58.03 + fastqc/0.74 | 97 | 15.52 | 6.65 | 27.46 | 24.31 + metaspades/3.15.3 | 6 | 4.56 | 7.46 | 28.00 | 26.65 + raxml/8.2.12 | 406 | 133.41 | 0.44 | 59.20 | 0.75 + + By default, only jobs from the last 30 days, tools that were allocated 8 GB of memory or more and run + more than 5 times, and jobs that ran for at least 5 minutes are included, since old jobs and fast and + infrequent tools can add noise to the results. + + You can change the age of jobs returned by specifying '--newer-than='. This option takes a value in the + PostgreSQL date/time interval format, see documentation: + https://www.postgresql.org/docs/current/functions-datetime.html + + Be sure to quote intervals containing spaces. + + The minimum job cutoff per tool can be specified with '--min-job-count', minimum runtime (in seconds) + with '--min-runtime', and minimum memory (in GB) with '--min-mem'. + EOF + + fields="run_count=1;total_runtime_hrs=2;avg_mem_used_gb=3;avg_mem_allocated_gb=4;avg_mem_percent=5" + tags="tool_id=0" + + read -r -d '' QUERY <<-EOF + WITH job_metrics AS ( + SELECT + j.id, + j.tool_id, + (SELECT metric_value FROM job_metric_numeric WHERE job_id = j.id AND metric_name = 'galaxy_memory_mb' LIMIT 1) * POW(1024, 2) AS mem_allocated, + (SELECT metric_value FROM job_metric_numeric WHERE job_id = j.id AND metric_name IN ('memory.peak', 'memory.memsw.max_usage_in_bytes') LIMIT 1) AS mem_used, + (SELECT metric_value FROM job_metric_numeric WHERE job_id = j.id AND metric_name = 'runtime_seconds' LIMIT 1) AS runtime, + (SELECT metric_value FROM job_metric_numeric WHERE job_id = j.id AND metric_name IN ('memory.peak', 'memory.memsw.max_usage_in_bytes') LIMIT 1) + / NULLIF((SELECT metric_value FROM job_metric_numeric WHERE job_id = j.id AND metric_name = 'galaxy_memory_mb' LIMIT 1) * POW(1024, 2), 0) AS mem_ratio + FROM + job j + WHERE + j.update_time > timezone('UTC', now()) - '$arg_newer_than'::INTERVAL + AND j.state = 'ok' + ) + SELECT + tool_id, + COUNT(*) AS run_count, + ROUND((SUM(runtime) / POW(60, 2))::numeric, 2) AS total_runtime_hrs, + ROUND((AVG(mem_used) / POW(1024, 3))::numeric, 2) AS avg_mem_used_gb, + ROUND((AVG(mem_allocated) / POW(1024, 3))::numeric, 2) AS avg_mem_allocated_gb, + ROUND((AVG(mem_ratio) * 100)::numeric, 2) AS avg_mem_percent + FROM + job_metrics + WHERE + mem_allocated >= ($arg_min_mem * POW(1024, 3)) + AND mem_used IS NOT NULL + AND runtime > $arg_min_runtime + GROUP BY + tool_id + HAVING + COUNT(tool_id) > $arg_min_job_count + ORDER BY + avg_mem_percent DESC + EOF +}