InfluxDB runtime

InfluxDB provides Go runtime profiles, trace, and other information useful for analyzing and debugging the server runtime execution.

Overview of Go runtime profiles

A Go runtime profile is a collection of stack traces showing call sequences that led to instances of a particular event. InfluxDB provides profile data for the following events:

  • blocks
  • CPU usage
  • memory allocation
  • mutual exclusion (mutex)
  • OS thread creation

When you send a profile request to InfluxDB, the Golang runtime pprof package samples the events on the runtime to collect stack traces and statistics (e.g., number of bytes of memory for heap allocation events). For some profiles, you can set the number of seconds that InfluxDB will collect profile data.

Once data collection is complete, InfluxDB returns the profile data. The default response format is a compressed protocol buffer in profile.proto format. profile.proto files are compatible with the pprof and go tool pprof analysis tools. For some profiles, InfluxDB provides an alternative human-readable plain text format with comments that translate to function calls and line numbers, but the pprof tools and profile.proto format offer the following advantages:

  • Read profiles from disk or HTTP.
  • Aggregate and compare multiple profiles of the same type.
  • Analyze and filter profile data.
  • Generate visualizations and reports.

Analyze Go runtime profiles

Use the /debug/pprof InfluxDB endpoints to download all the profiles at once or request them individually.

Get all runtime profiles

To download all runtime profiles at once, use an HTTP client to send a GET request to the /debug/pprof/all endpoint. go tool pprof can’t fetch profiles directly from /debug/pprof/all.

  1. GET http://localhost:8086/debug/pprof/all

InfluxDB returns a gzipped tar file that contains the following profiles in the profile.proto format:

OptionInclude by
Profile CPUPass a duration of seconds with the cpu query parameter in your request URL

Use an HTTP client like curl or wget to download profiles from /debug/pprof/all.

Example

  1. # Use `curl` to download a `.tar.gz` of all profiles after 10 seconds of CPU sampling.
  2. # Use `tar` to extract the profiles folder.
  3. curl "http://localhost:8086/debug/pprof/all?cpu=10s" | tar -xz
  4. # Analyze an extracted profile.
  5. go tool pprof profiles/heap.pb.gz

Profile all memory allocations

Profiles memory allocations and sets the default profile display to alloc_space, the total number of bytes allocated since the program began (including garbage-collected bytes).

  1. GET http://localhost:8086/debug/pprof/allocs
OptionInclude by
Seconds to samplePass an unsigned integer with the seconds query parameter in your request URL
Output plain text (mutually exclusive with seconds)Pass 1 with the debug query parameter in your request URL
  1. # Analyze the profile in interactive mode.
  2. go tool pprof http://localhost:8086/debug/pprof/allocs
  3. # `pprof` returns the following prompt:
  4. # Entering interactive mode (type "help" for commands, "o" for options)
  5. # (pprof)
  6. # At the prompt, get the top N memory allocations.
  7. (pprof) top10

Profile blocking operations

Profiles operations that led to blocking on synchronization primitives and caused Go to suspend goroutine’s execution.

  1. GET http://localhost:8086/debug/pprof/block
OptionInclude by
Output plain textPass 1 with the debug query parameter in your request URL
  1. # Analyze the profile in interactive mode.
  2. go tool pprof http://localhost:8086/debug/pprof/block
  3. # `pprof` returns the following prompt:
  4. # Entering interactive mode (type "help" for commands, "o" for options)
  5. # (pprof)
  6. # At the prompt, get the top N entries.
  7. (pprof) top10

Profile CPU

Profiles program counters sampled from the execution stack. To download the profile, use an HTTP client to send a GET request to the /debug/pprof/profile endpoint. go tool pprof can’t fetch the CPU profile directly.

  1. GET http://localhost:8086/debug/pprof/profile
OptionInclude by
Seconds to sample (default 30)Pass an unsigned integer with the seconds query parameter in your request URL

Use an HTTP client like curl or wget to download the profile.

Example

  1. # Get the profile.
  2. curl http://localhost:8086/debug/pprof/profile -o cpu
  3. # Analyze the profile in interactive mode.
  4. go tool pprof ./cpu
  5. # At the prompt, get the top N functions most often running
  6. # or waiting during the sample period.
  7. (pprof) top10

Use the seconds query parameter to control the sampling duration.

/debug/pprof/profile?seconds=SECONDS returns the same CPU profile as /debug/pprof/all?cpu=DURATION.

Example

  1. # Get the CPU profile after 10 seconds of sampling.
  2. curl "http://localhost:8086/debug/pprof/profile?seconds=10" -o cpu
  3. # Get all profiles after 10 seconds of CPU sampling.
  4. curl "http://localhost:8086/debug/pprof/all?cpu=10s" -o all.tar.gz

Profile goroutines

Profiles all current goroutines.

  1. GET http://localhost:8086/debug/pprof/goroutine
OptionInclude by
Seconds to samplePass an unsigned integer with the seconds query parameter in your request URL
Output plain text (mutually exclusive with seconds)Pass 1 with the debug query parameter in your request URL

Example

  1. # Analyze the profile in interactive mode.
  2. go tool pprof http://localhost:8086/debug/pprof/goroutine
  3. # `pprof` returns the following prompt:
  4. # Entering interactive mode (type "help" for commands, "o" for options)
  5. # (pprof)
  6. # At the prompt, get the top N entries.
  7. (pprof) top10

Profile heap memory allocations

Profiles heap, or memory allocations for live objects.

  1. GET http://localhost:8086/debug/pprof/heap
OptionInclude by
Run garbage control before samplingPass 1 with the gc query parameter in your request URL
Seconds to samplePass an unsigned integer with the seconds query parameter in your request URL
Output plain text (mutually exclusive with seconds)Pass 1 with the debug query parameter in your request URL

Example

  1. # Analyze the profile in interactive mode.
  2. go tool pprof http://localhost:8086/debug/pprof/heap
  3. # `pprof` returns the following prompt:
  4. # Entering interactive mode (type "help" for commands, "o" for options)
  5. # (pprof)
  6. # At the prompt, get the top N memory-intensive nodes.
  7. (pprof) top10
  8. # pprof displays the list:
  9. # Showing nodes accounting for 142.46MB, 85.43% of 166.75MB total
  10. # Dropped 895 nodes (cum <= 0.83MB)
  11. # Showing top 10 nodes out of 143

Profile mutual exclusions (mutexes)

Profiles holders of contended mutual exclusions (mutexes).

  1. GET http://localhost:8086/debug/pprof/mutex
OptionInclude by
Seconds to samplePass an unsigned integer with the seconds query parameter in your request URL
Output plain text (mutually exclusive with seconds)Pass 1 with the debug query parameter in your request URL

Example

  1. # Analyze the profile in interactive mode.
  2. go tool pprof http://localhost:8086/debug/pprof/mutex
  3. # `pprof` returns the following prompt:
  4. # Entering interactive mode (type "help" for commands, "o" for options)
  5. # (pprof)
  6. # At the prompt, get the top N entries.
  7. (pprof) top10

Profile thread creation

Profiles operations that led to the creation of OS threads.

  1. GET http://localhost:8086/debug/pprof/threadcreate
OptionInclude by
Seconds to samplePass an unsigned integer with the seconds query parameter in your request URL
Output plain text (mutually exclusive with seconds)Pass 1 with the debug query parameter in your request URL

Example

  1. # Analyze the profile in interactive mode.
  2. go tool pprof http://localhost:8086/debug/pprof/threadcreate
  3. # `pprof` returns the following prompt:
  4. # Entering interactive mode (type "help" for commands, "o" for options)
  5. # (pprof)
  6. # At the prompt, get the top N entries.
  7. (pprof) top10

Analyze the Go runtime trace

To trace execution events for InfluxDB, use the /debug/pprof/trace endpoint with go tool trace.

  1. GET http://localhost:8086/debug/pprof/trace

Example

  1. # Download the trace file.
  2. curl http://localhost:8086/debug/pprof/trace -o trace.out
  3. # Analyze the trace.
  4. go tool trace ./trace.out

Generate a pprof-like profile from trace

You can use go tool trace to generate pprof-like profiles from a trace file and then analyze them with go tool pprof.

Example

  1. # Generate a profile from the downloaded trace file.
  2. go tool trace -pprof=PROFILE_TYPE ./trace.out > PROFILE_TYPE.pprof

Replace PROFILE_TYPE with one of the following Golang profile types:

  • net: network blocking profile
  • sync: synchronization blocking profile
  • syscall: syscall blocking profile
  • sched: scheduler latency profile

View the command line that invoked InfluxDB

To view the command, arguments, and command-line variables that invoked InfluxDB, use the /debug/pprof/cmdline endpoint.

  1. GET http://localhost:8086/debug/pprof/cmdline

/debug/pprof/cmdline returns the command line invocation in plain text.

View runtime configuration

In InfluxDB v2.2+, you can view your active runtime configuration, including flags and environment variables. See how to view your runtime server configuration.