debian-mirror-gitlab/doc/administration/troubleshooting/log_parsing.md
2022-08-13 15:12:31 +05:30

8.7 KiB

stage group info
Systems Distribution To determine the technical writer assigned to the Stage/Group associated with this page, see https://about.gitlab.com/handbook/engineering/ux/technical-writing/#assignments

Parsing GitLab logs with jq (FREE SELF)

We recommend using log aggregation and search tools like Kibana and Splunk whenever possible, but if they are not available you can still quickly parse GitLab logs in JSON format (the default in GitLab 12.0 and later) using jq.

NOTE: Specifically for summarizing error events and basic usage statistics, the GitLab Support Team provides the specialised fast-stats tool.

What is JQ?

As noted in its manual, jq is a command-line JSON processor. The following examples include use cases targeted for parsing GitLab log files.

Parsing Logs

The examples listed below address their respective log files by their relative Omnibus paths and default filenames. Find the respective full paths in the GitLab logs sections.

General Commands

Pipe colorized jq output into less

jq . <FILE> -C | less -R

Search for a term and pretty-print all matching lines

grep <TERM> <FILE> | jq .

Skip invalid lines of JSON

jq -cR 'fromjson?' file.json | jq <COMMAND>

By default jq errors out when it encounters a line that is not valid JSON. This skips over all invalid lines and parses the rest.

Print a JSON log's time range

cat log.json | (head -1; tail -1) | jq '.time'

Use zcat if the file has been rotated and compressed:

zcat @400000006026b71d1a7af804.s | (head -1; tail -1) | jq '.time'

zcat some_json.log.25.gz | (head -1; tail -1) | jq '.time'

Get activity for correlation ID across multiple JSON logs in chronological order

grep -hR <correlationID> | jq -c -R 'fromjson?' | jq -C -s 'sort_by(.time)'  | less -R

Parsing gitlab-rails/production_json.log and gitlab-rails/api_json.log

Find all requests with a 5XX status code

jq 'select(.status >= 500)' <FILE>

Top 10 slowest requests

jq -s 'sort_by(-.duration_s) | limit(10; .[])' <FILE>
grep <PROJECT_NAME> <FILE> | jq .

Find all requests with a total duration > 5 seconds

jq 'select(.duration_s > 5000)' <FILE>

Find all project requests with more than 5 rugged calls

grep <PROJECT_NAME> <FILE> | jq 'select(.rugged_calls > 5)'

Find all requests with a Gitaly duration > 10 seconds

jq 'select(.gitaly_duration_s > 10000)' <FILE>

Find all requests with a queue duration > 10 seconds

jq 'select(.queue_duration_s > 10000)' <FILE>

Top 10 requests by # of Gitaly calls

jq -s 'map(select(.gitaly_calls != null)) | sort_by(-.gitaly_calls) | limit(10; .[])' <FILE>

Parsing gitlab-rails/production_json.log

Print the top three controller methods by request volume and their three longest durations

jq -s -r 'group_by(.controller+.action) | sort_by(-length) | limit(3; .[]) | sort_by(-.duration_s) | "CT: \(length)\tMETHOD: \(.[0].controller)#\(.[0].action)\tDURS: \(.[0].duration_s),  \(.[1].duration_s),  \(.[2].duration_s)"' production_json.log

Example output

CT: 2721   METHOD: SessionsController#new  DURS: 844.06,  713.81,  704.66
CT: 2435   METHOD: MetricsController#index DURS: 299.29,  284.01,  158.57
CT: 1328   METHOD: Projects::NotesController#index DURS: 403.99,  386.29,  384.39

Parsing gitlab-rails/api_json.log

Print top three routes with request count and their three longest durations

jq -s -r 'group_by(.route) | sort_by(-length) | limit(3; .[]) | sort_by(-.duration_s) | "CT: \(length)\tROUTE: \(.[0].route)\tDURS: \(.[0].duration_s),  \(.[1].duration_s),  \(.[2].duration_s)"' api_json.log

Example output

CT: 2472 ROUTE: /api/:version/internal/allowed   DURS: 56402.65,  38411.43,  19500.41
CT: 297  ROUTE: /api/:version/projects/:id/repository/tags       DURS: 731.39,  685.57,  480.86
CT: 190  ROUTE: /api/:version/projects/:id/repository/commits    DURS: 1079.02,  979.68,  958.21

Print top API user agents

jq --raw-output '[.route, .ua] | @tsv' api_json.log | sort | uniq -c | sort -n

Example output:

  89 /api/:version/usage_data/increment_unique_users  # plus browser details
 567 /api/:version/jobs/:id/trace       gitlab-runner # plus version details
1234 /api/:version/internal/allowed     GitLab-Shell

This sample response seems normal. A custom tool or script might be causing a high load if the output contains many:

You can also use fast-stats top to extract performance statistics.

Parsing gitlab-workhorse/current

Print top Workhorse user agents

jq --raw-output '[.uri, .user_agent] | @tsv' current | sort | uniq -c | sort -n

Example output:

  89 /api/graphql # plus browser details
 567 /api/v4/internal/allowed   GitLab-Shell
1234 /api/v4/jobs/request       gitlab-runner # plus version details

Similar to the API ua data, deviations from this common order might indicate scripts that could be optimized.

The performance impact of runners checking for new jobs can be reduced by increasing the check_interval setting, for example.

Parsing gitlab-rails/geo.log

Find most common Geo sync errors

If the geo:status Rake task repeatedly reports that some items never reach 100%, the following command helps to focus on the most common errors.

jq --raw-output 'select(.severity == "ERROR") | [.project_path, .message] | @tsv' geo.log | sort | uniq -c | sort | tail

Parsing gitaly/current

Use the following examples to troubleshoot Gitaly.

Find all Gitaly requests sent from web UI

jq 'select(."grpc.meta.client_name" == "gitlab-web")' current

Find all failed Gitaly requests

jq 'select(."grpc.code" != null and ."grpc.code" != "OK")' current

Find all requests that took longer than 30 seconds

jq 'select(."grpc.time_ms" > 30000)' current

Print top ten projects by request volume and their three longest durations

jq --raw-output --slurp '
  map(
    select(
      ."grpc.request.glProjectPath" != null
      and ."grpc.request.glProjectPath" != ""
      and ."grpc.time_ms" != null
    )
  )
  | group_by(."grpc.request.glProjectPath")
  | sort_by(-length)
  | limit(10; .[])
  | sort_by(-."grpc.time_ms")
  | [
      length,
      .[0]."grpc.time_ms",
      .[1]."grpc.time_ms",
      .[2]."grpc.time_ms",
      .[0]."grpc.request.glProjectPath"
    ]
  | @sh' current \
| awk 'BEGIN { printf "%7s %10s %10s %10s\t%s\n", "CT", "MAX DURS", "", "", "PROJECT" }
  { printf "%7u %7u ms, %7u ms, %7u ms\t%s\n", $1, $2, $3, $4, $5 }'

Example output

   CT    MAX DURS                              PROJECT
  206    4898 ms,    1101 ms,    1032 ms      'groupD/project4'
  109    1420 ms,     962 ms,     875 ms      'groupEF/project56'
  663     106 ms,      96 ms,      94 ms      'groupABC/project123'
  ...

Find all projects affected by a fatal Git problem

grep "fatal: " current | \
    jq '."grpc.request.glProjectPath"' | \
    sort | uniq

Parsing gitlab-shell/gitlab-shell.log

For investigating Git calls via SSH, from GitLab 12.10.

Find the top 20 calls by project and user:

jq --raw-output --slurp '
  map(
    select(
      .username != null and
      .gl_project_path !=null
    )
  )
  | group_by(.username+.gl_project_path)
  | sort_by(-length)
  | limit(20; .[])
  | "count: \(length)\tuser: \(.[0].username)\tproject: \(.[0].gl_project_path)" ' \
  gitlab-shell.log

Find the top 20 calls by project, user, and command:

jq --raw-output --slurp '
  map(
    select(
      .command  != null and
      .username != null and
      .gl_project_path !=null
    )
  )
  | group_by(.username+.gl_project_path+.command)
  | sort_by(-length)
  | limit(20; .[])
  | "count: \(length)\tcommand: \(.[0].command)\tuser: \(.[0].username)\tproject: \(.[0].gl_project_path)" ' \
  gitlab-shell.log