Fix another timeout when searching for pipelines

When we consider 'all' pipelines for MRs, we now mean:

1. The last 10,000 commits (unordered).
2. From the last 100 MR versions (newest first).

This seems to fix the MRs that time out on GitLab.com.

app/models/merge_request.rb

@@ -865,7 +865,19 @@ class MergeRequest < ActiveRecord::Base
   #
   def all_commit_shas
     if persisted?
-      column_shas = MergeRequestDiffCommit.where(merge_request_diff: merge_request_diffs).limit(10_000).pluck('sha')
+      # MySQL doesn't support LIMIT in a subquery.
+      diffs_relation =
+        if Gitlab::Database.postgresql?
+          merge_request_diffs.order(id: :desc).limit(100)
+        else
+          merge_request_diffs
+        end
+
+      column_shas = MergeRequestDiffCommit
+                      .where(merge_request_diff: diffs_relation)
+                      .limit(10_000)
+                      .pluck('sha')
+
       serialised_shas = merge_request_diffs.where.not(st_commits: nil).flat_map(&:commit_shas)
       (column_shas + serialised_shas).uniq

changelogs/unreleased/fix-502-mrs-with-lots-of-versions.yml

+---
+title: Ensure merge requests with lots of version don't time out when searching for
+  pipelines
+merge_request:
+author:
+type: performance