Database index corruption
How to recover from database index corruption.
A somewhat common configuration issue can lead to index corruption throughout the database. This page attempts to explain why this may occur and how to fix it.
Locale data and collation
Textual values in the database, such as usernames, or toot identifiers, are compared using so-called collation rules defining how characters are ordered and how to change their case. When setting up a database, Mastodon will use the database server’s default locale settings, including the default collation rules, which often is defined by the operating system’s settings.
Unfortunately, in late 2018, a glibc
update changed the collation rules for many locales, which means databases using an affected locale would now order textual values differently.
Since the database indexes are algorithmic structures which rely on the ordering of the values they are indexing, some of them would become inconsistent.
More information: https://wiki.postgresql.org/wiki/Locale_data_changes https://postgresql.verite.pro/blog/2018/08/27/glibc-upgrade.html
Am I affected by this issue?
If your database is not using C
or POSIX
for its collation setting (which you can check with SELECT datcollate FROM pg_database WHERE datname = current_database();
),
your indexes might be inconsistent, if you ever ran with a version of glibc prior to 2.28 and did not immediately reindex your databases after updating to glibc 2.28 or newer.
You can check whether your indexes are consistent using PostgreSQL’s amcheck
module: as the database server’s super user, connect to your Mastodon database and issue the following (this may take a while):
CREATE EXTENSION IF NOT EXISTS amcheck;
SELECT bt_index_check(c.oid)
FROM pg_index i
JOIN pg_class c ON i.indexrelid = c.oid
WHERE c.relname IN ('index_account_domain_blocks_on_account_id_and_domain',
'index_account_proofs_on_account_and_provider_and_username',
'index_accounts_on_username_and_domain_lower', 'index_accounts_on_uri',
'index_accounts_on_url', 'index_conversations_on_uri',
'index_custom_emoji_categories_on_name',
'index_custom_emojis_on_shortcode_and_domain',
'index_devices_on_access_token_id', 'index_domain_allows_on_domain',
'index_domain_blocks_on_domain', 'index_email_domain_blocks_on_domain',
'index_invites_on_code', 'index_markers_on_user_id_and_timeline',
'index_media_attachments_on_shortcode', 'index_preview_cards_on_url',
'index_statuses_on_uri', 'index_tags_on_name_lower',
'index_tombstones_on_uri', 'index_unavailable_domains_on_domain',
'index_users_on_email', 'index_webauthn_credentials_on_external_id'
);
If this raises an error, your database is corrupted and needs fixing. If it does not, you may need to perform more involved checks to be sure. Unlike the previous checks, those more involved checks will lock tables when running, thus interfering with the availability of your instance.
CREATE EXTENSION IF NOT EXISTS amcheck;
SELECT bt_index_parent_check(c.oid)
FROM pg_index i
JOIN pg_class c ON i.indexrelid = c.oid
WHERE c.relname IN ('index_account_domain_blocks_on_account_id_and_domain',
'index_account_proofs_on_account_and_provider_and_username',
'index_accounts_on_username_and_domain_lower', 'index_accounts_on_uri',
'index_accounts_on_url', 'index_conversations_on_uri',
'index_custom_emoji_categories_on_name',
'index_custom_emojis_on_shortcode_and_domain',
'index_devices_on_access_token_id', 'index_domain_allows_on_domain',
'index_domain_blocks_on_domain', 'index_email_domain_blocks_on_domain',
'index_invites_on_code', 'index_markers_on_user_id_and_timeline',
'index_media_attachments_on_shortcode', 'index_preview_cards_on_url',
'index_statuses_on_uri', 'index_tags_on_name_lower',
'index_tombstones_on_uri', 'index_unavailable_domains_on_domain',
'index_users_on_email', 'index_webauthn_credentials_on_external_id'
);
If this succeeds, without returning an error, your database should be consistent, and you can safely disregard the warning Mastodon emits when running db:migrate
.
Fixing the issue
Unless you take action, if you are affected, your database could get more and more inconsistent as the time pass. Therefore, it is important to fix it as soon as possible.
Mastodon 3.2.2 and later come with a semi-interactive script to fix those corruptions as best as possible. If you’re on an earlier version, update to 3.2.2 first. It is possible that running the database migrations to 3.2.2 will fail because of those very corruptions, but the database should then be brought to a state that the maintenance tool bundled with Mastodon 3.2.2 can then recover from.
Before attempting to fix your database, stop Mastodon and make a backup of your database. Then, with Mastodon still stopped, run the maintenance script:
RAILS_ENV=production tootctl maintenance fix-duplicates
The tool will walk through the database to find duplicate and fix them. In some cases, those operations are destructive. In the most destructive cases, you will be asked to chose which record to keep. In all cases, walking through the whole database in search of duplicates is an extremely long operation. Mastodon has to be stopped during the whole process to prevent additional duplicates from occuring.
Avoiding the issue
To avoid the issue, reindex your database immediately after any libc update.
The
SQL command REINDEX
or the
reindexdb
command-line tool
may be useful for this.
Last updated August 29, 2021 · Improve this page