Technologies/Langfuse/langfuse.container.cpu.utilization
LangfuseLangfuseMetric

langfuse.container.cpu.utilization

Container CPU utilization percentage
Dimensions:None
Available on:CloudWatchCloudWatch (1)
Interface Metrics (1)
CloudWatchCloudWatch
CPU utilization percentage for the container
Dimensions:None

Technical Annotations (27)

Configuration Parameters (11)
LANGFUSE_INGESTION_QUEUE_PROCESSING_CONCURRENCYrecommended: lower than default
Controls parallel job processing; lowering reduces memory spikes at cost of throughput
LANGFUSE_INGESTION_CLICKHOUSE_WRITE_BATCH_SIZErecommended: lower than default
Controls data batched per write; lowering reduces memory consumption
LANGFUSE_TOKEN_COUNT_WORKER_POOL_SIZErecommended: lower than 2 (default)
Tokenizer thread pool; high values cause pending requests to accumulate in memory
NODE_OPTIONSrecommended: --max-old-space-size=6144
Caps Node.js heap to container memory limit, prevents runaway usage
web.startupProbe.initialDelaySecondsrecommended: 60
Allows time for initial migrations before first health check
web.livenessProbe.initialDelaySecondsrecommended: 120
Prevents premature restart during migration completion
web.readinessProbe.initialDelaySecondsrecommended: 120
Delays readiness check until migrations complete
web.startupProbe.periodSecondsrecommended: 15
Check interval during startup phase
web.livenessProbe.periodSecondsrecommended: 15
Check interval for liveness
web.startupProbe.failureThresholdrecommended: 5
Number of failures before pod restart
container_versionrecommended: v3.x
V3 contains major performance improvements to address CPU spikes
Error Signatures (3)
context deadline exceeded (Client.Timeout exceeded while awaiting headers)log pattern
read: connection reset by peerlog pattern
CrashLoopBackOfferror code
CLI Commands (3)
ps -eo pid,comm,args | awk "/[n]ode( |$)/ && !/node -e/ {print}"diagnostic
kill -USR1 $PIDdiagnostic
helm upgrade -i langfuse -f langfuse/langfuse -n observability-backend-langfuse-dev --version 0.13.1remediation
Technical References (10)
BullMQcomponentNode.js heapconceptstartupProbecomponentlivenessProbecomponentreadinessProbecomponent/api/public/readyfile path/api/public/healthfile pathtiktokencomponent/api/public/ingestionendpointbatch processingconcept
Related Insights (6)
Worker container high memory usage with low CPU under ingestion loadcritical
Web container high memory usage with low CPU utilizationwarning
Langfuse web pod fails health checks during startupcritical
CPU spikes cause Langfuse container crashes and API unavailabilitycritical
Tiktoken tokenization causes high CPU usage on large trace inputswarning
Ingestion endpoint requires more backend resources than other API endpointswarning