rc_id will hit its max at 2147483647 rather than 4294967295 because it's an int rather than an unsigned int. Therefore, it should be changed to unsigned.
Maybe this is also a good opportunity to change it to an unsigned bigint, since it's at 600 million already on enwiki and at the current rate of increase could reach its limit in our lifetime.
| Subject | Repo | Branch | Lines +/- | |
|---|---|---|---|---|
| Make rc_id unsigned | mediawiki/core | master | +629 -1 |
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Open | Feature | None | T63111 Convert primary key integers and references thereto from int to bigint (unsigned) | ||
| Resolved | Ladsgroup | T62962 The primary key of recentchanges (rc_id) table should be unsigned | |||
| Resolved | Umherirrender | T77005 Database field user_newtalk.user_id is not the same data type as user.user_id |
Some fields that should be unsigned, that aren't:
*change_tag.ct_rc_id
*change_tag.ct_log_id
*change_tag.ct_rev_id
*filearchive.fa_id
*filearchive.fa_deleted_user
*ipblocks.ipb_id
*page_props.pp_page
*page_restrictions.pr_page
*recentchanges.rc_id
*tag_summary.ts_rc_id
*tag_summary.ts_log_id
*tag_summary.ts_rev_id
*user_newtalk.user_id
*user_properties.up_user
Change 112825 had a related patch set uploaded by MZMcBride:
Make the following fields int unsigned:
@Umherirrender has been doing some of these; see T77005 and Gerrit topic "schemaunsigned".
In the current MediaWiki master (d056d4c2), there are no signed fields for any primary keys.
rc_id is still signed as far as I can tell – at least, it’s not explicitly unsigned in the documentation, [in maintenance/tables.sql](https://gerrit.wikimedia.org/r/plugins/gitiles/mediawiki/core/+/d056d4c273a069d572ceb2160346ee697d7b87de/maintenance/tables.sql#1377), on the replicas, or on mwdebug1002. Or am I misunderstanding something? (I’m not very familiar with (My)SQL, sorry, but as far as I can tell from a quick Google, signed is the default.)
@LucasWerkmeister Yeah, I checked the schema for any explicit use of signed, and didn't think about the implicit ones until I saw T63111.
As a generic task, these are basically the same, but rather than merging it, perhaps we can keep this as sub-task focussed only on the rc_id field which is the most pressing one, and might make sense to adopt a non-bigint solution for first.
I am fine going to unsigned for rc_id . Once tables.sql is merged with this change, just add Schema-change-in-production as a tag (https://wikitech.wikimedia.org/wiki/Schema_changes#Workflow_of_a_schema_change) and we can take care of it :)
I agree with start working in this soon with high priority, rc_id will run out of range "soon" (worst auto_increment to have into account). Some stats:
I will do this as part of migrating the rc table to abstract schema (T230428: Migrate tables.sql to abstract schema)
Change 667356 had a related patch set uploaded (by Ladsgroup; owner: Ladsgroup):
[mediawiki/core@master] Make rc_id unsigned
The code-wise work for the schema change is done. Production schema change is being tracked in T276150: Schema change to make rc_id unsigned and rc_timestamp BINARY.