Skip to content

Latest commit

 

History

History
29 lines (20 loc) · 3.77 KB

MORE_INFO.md

File metadata and controls

29 lines (20 loc) · 3.77 KB
Raw

Syncback: More Info

This doc contains low-level details about the Syncback extension. Most developers won't need to touch this doc, but it's here to provide additional information if you're trying to debug something or are just curious. Still having trouble? Let us know!

Gotchas

Tilt updates my pod after a syncback

If fileX changes on the remote container and your syncback task copies fileX back to your local machine, then depending on your project setup, Tilt may register this file change and rebuild the associated resource. This is expected behavior--there's no good way to determine whether a file changed because it was edited locally, or if it was synced back from a container.

If you're using Live Update, we expect the update that runs after syncback to be a no-op, and super fast. If this extra update is bothering you, here are some things that might help:

  1. if you're not already using Live Update, start; instead of a whole image rebuild, Tilt will just copy over the changed files, and you'll save lots of time.
  2. if you're already using Live Update but it's still too slow, consider making use of run triggers to ensure that your run steps only execute when they need to (e.g. if you run a go build as part of your Live Update, use triggers to make sure it only runs when packages in ./internal change).
  3. if you never want fileX to trigger a Tilt update, ignore it with .tiltignore or watch_settings(ignore=) (see docs).

Syncback isn't finding the files I expect it to

Kubernetes might have connected you to the wrong pod. If you're passing any k8s_object besides a pod ID (e.g. deploy/my-deploy), the syncback will run on the first pod Kubernetes finds belonging to that object. If you run the syncback at a time when there are old pods hanging around--e.g. right after Tilt deployed a new pod, the old pod might take a few seconds to start shutting down--then you risk running the syncback on an out-of-date pod. We recommend waiting a few seconds/waiting until kubectl get pods shows only the pods you expect to, and trying again.

Notes on Implementation

How We Pick Pods

Under the hood, this script uses kubectl exec to run rsync (and some other assorted commands). As detailed here, you may call kubectl exec on a specific pod, or on any Kubernetes object specified as TYPE/NAME; in the latter case, the command executed on the first pod belonging to that object.

About Syncing

Under the hood, this extension uses rsync. Rsync, which stands for "remote sync", is a remote and local file synchronization tool. It uses an algorithm that minimizes the amount of data copied by only moving the portions of files that have changed.[^1]

rsync isn't the simplest implementation here--it might have been easier to run kubectl cp <POD>:/src_dir ./target_dir, for instance. We chose rsync because:

  • it handles deletions and file renames, which the the kubectl cp solution above does not
  • it's more efficient: if only 2MB of files have changed within a 2GB directory, you want to copy down only what has changed, and avoid moving multiple gigs of data

This extension uses the krsync.sh script (adapted from Karl Bunch's elegant script on Server Fault) to bridge the gap between rsync and Kubernetes. As part of that script, we copy our pre-built rsync binary onto the container, and use that binary for all subsequent rsync calls.