Skip to content

check-crash.md

Latest commit

 

History

History
152 lines (108 loc) · 4.23 KB

check-crash.md

File metadata and controls

152 lines (108 loc) · 4.23 KB

groonga-query-log-check-crash

groonga-query-log-check-crash is a tool that supports recovery when Groonga crashes.

If Groonga crashes while updating a database, the database may become corrupted. It shows tables, columns, and indexes that require recovery, so you can use it when checking what needs to be recovered.

Usage

This section describes how to use groonga-query-log-check-crash.

You need both the process log and the query log to run groonga-query-log-check-crash. For log output settings and content, see the Groonga documentation: https://groonga.org/docs/reference/log.html

Specify the process log and query log. There is no fixed order for specifying them, so specify all logs.

Command example:

groonga-query-log-check-crash \
  path/to/process-log-2025-* \
  path/to/query-log-2025-* \
  path/to/process-log-2026-* \
  path/to/query-log-2026-*

Options

--command-format

This script displays any Groonga commands that terminated abnormally. Specify the format of that command.

You can specify command or uri, and output will be as follows.

--command-format=command:

===
[unflushed] 2000-01-01T00:00:01+09:00: 
load \
  --table "Data"
===

--command-format=uri:

===
[unflushed] 2000-01-01T00:00:01+09:00: /d/load?table=Data
===

The default is command. When using it as HTTP server, specifying uri makes it easier to use when re-running.

--pretty-print

This is only available when command is specified in --command-format.

By default, --pretty-print is enabled. Disabling it will output Groonga commands on a single line.

--no-pretty-print:

[unflushed] 2000-01-01T00:00:01+09:00: load --table "Data"

--output-level

You can specify info or debug. The default value is info. Please use info as the default. debug provides more detailed output for developers.

Reading the Output

Summary

The Summary includes the following items.

  • crashed
    • In the case of yes, Groonga has crashed.
    • Please report it with the logs.
  • unflushed
    • In the case of yes, data in memory (non-persistent data) may not have been written to disk (persistent data).
    • Please re-run the target command.
  • unfinished
    • In the case of yes, the target table, column, or index may be broken.
    • Please rebuild the target table, column, or index.
  • leak
    • In the case of yes, Groonga is leaking memory.
    • Please report it with the logs.

Example:

Summary:
crashed:yes, unflushed:yes, unfinished:no, leak:no

Unflushed

If the target command exists, the following output is displayed.

!!!
!!! [unflushed] Recovery information
!!!
There may be commands that were not flushed between 2000-01-01T00:00:00+09:00 and 2000-01-01T12:00:00+09:00.
These commands may not have been written to the database files, so please re-run them.
===
[unflushed] 2000-01-01T00:00:01+09:00: load --table "Data"
===

Summary:
crashed:yes, unflushed:yes, unfinished:no, leak:no
NG: Please check the display and logs.

Lines beginning with [unflushed] are commands that may not have been flushed. Since it may not have been flushed, please re-run that command.

Unfinished

If the target command exists, the following output is displayed.

!!!
!!! [unfinished] Recovery information
!!!
Unfinished commands were found due to abnormal termination or other issues.
It is safer to rebuild the target tables, columns, and indexes because the data may be corrupted.
===
[unfinished] 2000-01-01T00:00:01+09:00: load --table "Data"
===

Summary:
crashed:yes, unflushed:no, unfinished:yes, leak:no
NG: Please check the display and logs.

Lines beginning with [unfinished] are commands that did not complete normally due to abnormal termination or other issues. In this case, Data and any dependent columns or indexes may be broken and need to be rebuilt.

Others

If any output other than unflushed or unfinished was displayed, there might have been an issue with Groonga. Please report the problem to the Groonga team using the links below, and include the logs used for the check if possible.

This will help with investigating the issue.