Crash on blob view due to display of last commit

Reported by @ced in the Matrix channel:

Very frequently browsing content file crashes like https://foss.heptapod.net/tryton/tryton/-/blob/branch/default/modules/stock_split/stock.py The request GET https://foss.heptapod.net/tryton/tryton/-/blob/branch/default/modules/stock_split/stock.py?format=json&viewer=simple returns 500

Corresponding abridged logs:

{
  "exception.class": "NoMethodError",
  "exception.message": "undefined method `sha' for nil:NilClass",
  "exception.backtrace": [
    "app/controllers/projects/blob_controller.rb:280:in `set_last_commit_sha'",
    "app/controllers/projects/blob_controller.rb:294:in `show_json'",
    "app/controllers/projects/blob_controller.rb:85:in `block (2 levels) in show'",
    "app/controllers/projects/blob_controller.rb:77:in `show'"]
}

Now the surprise is that it does not use the gRPC LastCommitForPath method, but FindCommits instead. lib/gitlab/git/commit.rb:

        def where(options)
          repo = options.delete(:repo)
          raise 'Gitlab::Git::Repository is required' unless repo.respond_to?(:log)

          repo.log(options)
        end

        # (...)

        def last_for_path(repo, ref, path = nil, literal_pathspec: false)
          # rubocop: disable Rails/FindBy
          # This is not where..first from ActiveRecord
          where(
            repo: repo,
            ref: ref,
            path: path,
            limit: 1,
            literal_pathspec: literal_pathspec
          ).first
          # rubocop: enable Rails/FindBy
        end

and of course, lib/gitlab/git/repository.rb:

      def log(options)
        default_options = {
          limit: 10,
          offset: 0,
          path: nil,
          author: nil,
          follow: false,
          skip_merges: false,
          after: nil,
          before: nil,
          all: false
        }

        options = default_options.merge(options)
        options[:offset] ||= 0

        limit = options[:limit]
        if limit == 0 || !limit.is_a?(Integer)
          raise ArgumentError, "invalid Repository#log limit: #{limit.inspect}"
        end

        wrapped_gitaly_errors do
          gitaly_commit_client.find_commits(options)
        end
      end

(all these files are identical in branches heptapod-oldstable and heptapod-0-41)

The FindCommits call itself in HGitaly logs:

191325:[2024-01-25 10:52:06,374] [correlation_id=01HN03Z4MMSPB90878C7YFT4VP] [15017] [INFO] [hgitaly.service.interceptors] Starting to process RPC FindCommitsRequest { repository { storage_name: "default" relative_path: "@hashed/0c/62/0c62cc42d6479a691f03083654ab6a7a84229ab156c948ba8d3b6c79ddd95536.git" gl_repository: "project-1049" gl_project_path: "tryton/tryton" } revision: "branch/default" limit: 1 paths: "modules/stock_split/stock.py" disable_walk: true global_options { literal_pathspecs: true } }

So it looks like FindCommits returned an empty list, on which .first returned nil, and the question is why.

Local hg calls with ::default or same which actual node id found in GitLab branches state file are correct:

hg log -r 'reverse(::7bbbb91621ab3759dfdc04de3aa1754c89ef145d)' modules/stock_split/stock.py 
changeset:   100125:184cee9c3825
user:        Cédric Krier <ced@b2ck.com>
date:        Sun Aug 20 12:17:27 2023 +0200
summary:     Normalize Unit of Measure labels

changeset:   100120:7129f417a22c
user:        Cédric Krier <ced@b2ck.com>
date:        Tue Aug 22 13:18:33 2023 +0200
summary:     Rename Uom to Unit for move and inventory