Skip to content

Latest commit

 

History

History
69 lines (58 loc) · 3.59 KB

advanced.md

File metadata and controls

69 lines (58 loc) · 3.59 KB
Raw

Advanced Features

Run Maintenance in the Foreground

Scalar runs maintenance on your repositories in the background. You could chose to run those maintenance steps yourself by running scalar run <task> [<options>]. This command will run one maintenance step as specified by <task>. These tasks are:

  • config: Set recommended Git config settings. These are all intended to improve performance in large repos. If the repo was cloned by Scalar, then most of these settings will overwrite any existing settings. Otherwise, Scalar will not override a local config value that disagrees with the recommendation.

  • commit-graph: Update the Git commit-graph to include all reachable commits. After writing a new file, verify the file was computed successfully. This drastically improves the performance of commands like git log --graph.

  • fetch: Fetch the latest data from the remote servers. If using the GVFS protocol, download the latest set of commit and tree packs from the cache server or the origin remote. Otherwise, this step will fetch the latest objects from each remote. This will not update your local refs, so your git fetch commands will report all ref updates. Those git fetch commands will be much faster as they will require much less data transfer.

  • loose-objects: Examine the loose objects contained in the shared object cache. First, delete any loose objects that are currently located in pack-files. Second, create a new pack-file containing the remaining loose objects. The pack will contain up to 50,000 objects, leaving any extra objects for the next run of the loose-objects step. This happens in the background once a day.

  • pack-files: Update the Git multi-pack-index and repack small pack-files into larger pack-files. Scalar downloads many pack-files during the fetch step, or during git checkout commands. The pack-files step updates the Git multi-pack-index to improve lookup speed. Further, it combines pack-files into larger files to reduce the total pack-file count. This step is designed to work without blocking concurrent processes by only deleting pack-files after they were marked as "unused" for at least a day. This step allows using the --batch-size=<size> option. By default, the batch-size is "2g" for two gigabytes. This batch size signifies the goal size of a repacked pack-file.

  • all: This task runs all of the above steps in the following order:

    1. config ensures our recommended values are set for the remaining steps.
    2. fetch downloads the latest data from the remotes.
    3. commit-graph updates based on the newly fetched data.
    4. loose-objects cleans up loose objects. When using the GVFS protocol, the previous steps may have downloaded new loose objects.
    5. pack-files cleans up the pack-files, including those downloaed in previous steps.

Controlling Background Maintenance

If you have a need to temporarily pause background maintenance from running, then you can run scalar pause [<number>] to stop all maintenance for <number> hours. If <number> is not provided, then maintenance will be delayed for 12 hours.

For example, you may want to pause maintenance if you need all CPU resources on your machine available for a high-performance activity or for performance testing. You may also want to pause maintenance if you are on a metered internet connection, as some of the maintenance downloads Git data from the remote server.