WIP: By-repository mutex to create issues

[guillep2k]

This should fix a possible race condition in models/issue.go (#7887) if we deem it necessary to fix; I'm not sure it's much of an issue, however.

The idea is to put a lock between select max(index) and insert into issue (index, ....), so there's no chance that two simultaneous requests attempt to insert the same index for two separate issues.

To make sure it doesn't hurt insert performance, I've used per-repository mutexes with a library I've found at go.chromium.org/luci/common/sync/mutexpool; so there's no lock if two inserts go for different repositories.

As I've said, it's probably not worth fixing; but I figured I'd try anyway. The pattern might be useful for other parts of the code.

One more thing: I couldn't figure out a good way of unit testing this modification. I've been looking around but didn't find anything convincing. Suggestions are welcomed.

[codecov-io]

Codecov Report

❗ No coverage uploaded for pull request base (master@4bc4acd). Click here to learn what that means.
The diff coverage is 69.23%.

@@            Coverage Diff            @@
##             master    #7894   +/-   ##
=========================================
  Coverage          ?   41.41%           
=========================================
  Files             ?      478           
  Lines             ?    63951           
  Branches          ?        0           
=========================================
  Hits              ?    26486           
  Misses            ?    34015           
  Partials          ?     3450

Impacted Files	Coverage Δ
models/issue.go	50% <69.23%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4bc4acd...e6bbc19. Read the comment docs.

[lunny]

@guillep2k I don't think this will work on a scale architecture i.e. gitea.com . There are serval gitea instances shared with databases and file system. This will not work.

I would like a simple but ugly fix to retry the creating when failed because of the max id are already exist.

[guillep2k]

@guillep2k I don't this will work on a scale architecture i.e. gitea.com . There are serval gitea instance shared with databases and file system. This will not work.

I would like a simple but ugly fix to retry the creating when failed because of the max id are already exist.

You are right. I'll close this PR and make another. I'll investigate how to detect duplicate key with xorm.