Technologies/NGINX/gunicorn.request.duration
NGINXNGINXMetric

gunicorn.request.duration

HTTP request processing duration
Dimensions:None

Technical Annotations (105)

Configuration Parameters (25)
timeoutrecommended: 120
Worker timeout in seconds; increase when requests take longer than default
workersrecommended: 9
For 4-core machine using formula (2 * CPUs) + 1
worker-classrecommended: gevent
Non-blocking worker type for high I/O concurrency in dashboards
threadsrecommended: 1
Threads less critical when using Gevent workers
http_request_duration_seconds.bucketsrecommended: [0.01, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10]
Histogram buckets for latency distribution enabling accurate percentile calculations
NEW_RELIC_CONFIG_FILErecommended: newrelic.ini
Path to New Relic agent configuration file for APM integration
scrape_configs.job_namerecommended: django_app
Identifier for the Django application scrape job in Prometheus
scrape_configs.static_configs.targetsrecommended: ['localhost:8000']
Target endpoint(s) for Prometheus to scrape Django/Gunicorn metrics
--bindrecommended: unix:/run/gunicorn.sock
Use Unix socket for local Nginx-Gunicorn communication instead of TCP (0.0.0.0:8000)
proxy_passrecommended: http://unix:/run/gunicorn.sock
Nginx should proxy to Gunicorn via Unix socket, not TCP localhost
ListenStreamrecommended: /run/gunicorn.sock
Systemd socket file configuration for Gunicorn socket activation
--timeoutrecommended: 60
Gunicorn worker timeout in seconds for handling requests
--workersrecommended: 3
Number of Gunicorn worker processes for handling concurrent requests
statsd_prefixrecommended: your_app_name
prefix for metrics sent to StatsD aggregator
statsd_hostrecommended: metrics-aggregator:9125
StatsD endpoint for pushing metrics from all workers
NUM_WORKERSrecommended: tested 1, 3, 9, 17 - no improvement
worker count does not resolve the CPU spike issue
accesslogrecommended: -
Log to stdout for monitoring worker utilization
access_log_formatrecommended: %(h)s %(l)s %(u)s %(t)s "%(r)s" %(s)s %(b)s "%(f)s" "%(a)s" rt=%(L)s busy=%({x-busy}i)s
Include request time (rt=%(L)s) and busy worker count for performance analysis
backlog
controls size of pending connection queue when all workers are busy
--access-logfilerecommended: '-'
Writes access logs to stdout for App Service log collection
--error-logfilerecommended: '-'
Writes error logs to stderr for App Service log collection
appendfsyncrecommended: no
Redis parameter; use 'no' instead of 'everysec' to prevent blocking disk writes
worker_tmp_dirrecommended: /dev/shm or tmpfs mount
Must use memory-backed filesystem to avoid blocking on os.fchmod
preload_apprecommended: True
Required in Gunicorn config when using custom Dockerfile on Cloud Run to eliminate timeout cycle
-wrecommended: 4 (or (2 × CPU cores) + 1)
Worker count; insufficient workers cause timeouts under load
Error Signatures (11)
[CRITICAL] WORKER TIMEOUT (pid:log pattern
500http status
[CRITICAL] WORKER TIMEOUTlog pattern
WORKER TIMEOUT (pid:log pattern
Worker (pid:*) was sent SIGKILL! Perhaps out of memory?log pattern
upstream prematurely closed connection while reading response header from upstreamlog pattern
502http status
H12 Request Timeouterror code
WORKER TIMEOUTlog pattern
504http status
Worker exiting (pid:log pattern
CLI Commands (14)
sudo systemctl daemon-reloadremediation
sudo systemctl restart gunicornremediation
gunicorn app:server --workers $WORKERS --worker-class gevent --bind 0.0.0.0:8000 --timeout 60remediation
pip install gunicorn geventremediation
pip install newrelicmonitoring
newrelic-admin generate-config YOUR_LICENSE_KEY newrelic.inimonitoring
NEW_RELIC_CONFIG_FILE=newrelic.ini newrelic-admin run-program gunicorn myapp.wsgi:applicationmonitoring
sudo systemctl restart nginxremediation
py-spy --subprocessesdiagnostic
--timeout 120remediation
free -mdiagnostic
topdiagnostic
--access-logfilediagnostic
-w 4remediation
Technical References (55)
workercomponentsystemdcomponentExecStartcomponentWSGIprotocolGeventcomponentDash callbackscomponentASGIprotocolEventletcomponentCelerycomponentP95concepthistogram_quantileconceptRED Methodconceptnewrelic-admincomponenttransaction tracesconceptscrape_configscomponentstatic_configscomponent/run/gunicorn.sockfile path/etc/systemd/system/gunicorn.socketfile path/etc/nginx/conf.d/filename.conffile pathUnix socketconceptmemcachedcomponentmanage.py runservercomponentconnection backlogconceptupstream_response_timecomponentmaster/child process modelconceptApp Service LinuxcomponentCProfilecomponentSNAT port exhaustionconceptos.fchmodconceptheartbeatcomponentfaulthandlercomponentworker processcomponentGoogle App EnginecomponentGoogle Cloud RuncomponentB2 instance classcomponentgthread workercomponentHTTP request smugglingconceptreverse proxycomponentmaster processcomponenttmpfscomponenthttp.disconnectcomponentasyncio.CancelledErrorexceptionasyncio.wait_forcomponentdatabase queriesconceptindexesconceptload balancercomponenthealth checkconceptgeventcomponentuvicorncomponentsocket accept queueconceptbacklogconceptsocket queueconceptswap memoryconceptgunicorn workercomponentDjango Debug Toolbarcomponent
Related Insights (39)
Gunicorn worker timeout on long-running requestswarning
▸
Worker pool exhaustion causes exact 60-second request hangscritical
▸
Synchronous WSGI workers block on I/O preventing concurrent request handlingwarning
▸
Request timeout rate increases with each load testwarning
▸
Slow backend responses cause concurrent request memory buildupwarning
▸
P95 latency exceeding 2 seconds degrades user experiencewarning
▸
Elevated request duration indicates performance degradationwarning
▸
Gunicorn startup with New Relic APM wrapperinfo
▸
Prometheus metrics scraping from Django applicationinfo
▸
TCP socket binding causes severe request latency compared to Unix socketcritical
▸
Insufficient Gunicorn workers cause unpredictable response timeswarning
▸
Multi-process model complicates Prometheus metrics collectioninfo
▸
Long-running requests terminated by worker timeoutwarning
▸
Gunicorn CPU spike to 100% causes severe page load delayscritical
▸
Request queueing in connection backlog due to insufficient workerswarning
▸
Requests queue in connection backlog when workers are saturatedwarning
▸
Worker killed when timeout setting too low or missingcritical
▸
Worker timeout due to high CPU causing slow request processingwarning
▸
Worker timeout from long running requests exceeding timeout thresholdwarning
▸
Redis appendfsync blocking causes Gunicorn worker timeoutcritical
▸
Worker heartbeat blocks indefinitely on disk-backed filesystemcritical
▸
Worker timeout kills process without logging request URIwarning
▸
Slow external dependencies cause worker saturation and cascading failureswarning
▸
Gunicorn workers enter infinite timeout-SIGKILL cycle on Google App Enginecritical
▸
Request interpretation desynchronization between proxy and backendcritical
▸
Master process disk I/O blocking causes multi-second response latencycritical
▸
Request persistence after platform timeout causes resource wastewarning
▸
Worker timeout kills unresponsive workers after 30 secondscritical
▸
Database bottlenecks cause worker timeoutswarning
▸
Frequent health checks trigger unnecessary worker killswarning
▸
External service delays trigger worker timeoutswarning
▸
Slow application logic causes worker timeoutswarning
▸
Request duration metric excludes socket accept queue timewarning
▸
Socket backlog queue time not exposed as metricinfo
▸
Request duration metric excludes socket queue wait timewarning
▸
Memory swapping causes slow response times and Nginx 504 timeoutscritical
▸
Single Gunicorn worker insufficient for production without benchmarkingwarning
▸
Gunicorn worker timeout causes 20-30 second page delayscritical
▸
Large gap between Django CPU time and total request time indicates WSGI layer bottleneckinfo
▸