Files @ ad374c171656
Branch filter:

Location: kallithea/docs/usage/locking.rst

ad374c171656 1.3 KiB text/prs.fallenstein.rst Show Annotation Show as Raw Download as Raw
Mads Kiilerich
comments: introduce lazy join of ChangesetStatus on ChangesetComment

99% of all uses of ChangesetComment will also need the optional corresponding
ChangesetStatus (if any) on the status_change relationship. Fetching it on
demand gives a lot of roundtrips and might be slow ... but adding explicit bulk
queries everywhere do not seem feasible. Adding lazy=joined in the data model
seems like the best solution.

Loading a pull request with 10 comments on a slow repo goes from 3.5 s to
2.7 s.
.. _locking:

===================================
Kallithea repository locking system
===================================


| Repos with **locking function=disabled** is the default, that's how repos work
  today.
| Repos with **locking function=enabled** behaves like follows:

Repos have a state called `locked` that can be true or false.
The hg/git commands `hg/git clone`, `hg/git pull`, and `hg/git push`
influence this state:

- The command `hg/git pull <repo>` will lock that repo (locked=true)
  if the user has write/admin permissions on this repo

- The command `hg/git clone <repo>` will lock that repo (locked=true) if the
  user has write/admin permissions on this repo


Kallithea will remember the user id who locked the repo
only this specific user can unlock the repo (locked=false) by calling

- `hg/git push <repo>`

every other command on that repo from this user and
every command from any other user will result in http return code 423 (locked)


additionally the http error includes the <user> that locked the repo
(e.g. “repository <repo> locked by user <user>”)


So the scenario of use for repos with `locking function` enabled is that
every initial clone and every pull gives users (with write permission)
the exclusive right to do a push.


Each repo can be manually unlocked by admin from the repo settings menu.