Capistrano deletes last working deployment if recovering from too many failures

[tminich]

Steps to reproduce

• Have X or more deployments fail (X being keep_releases-1)
• Have a successful deployment

Expected behavior

Capistrano keeps (at least) the last known working release

Actual behavior

Last known working release gets deleted and failed releases are kept so there is no working release to roll back to if necessary

Arguably keep_releases should apply to successful and failed deployments separately (or there should be separate configuration options for these cases). I went through the code a bit and from what I can see (not very familiar with Ruby) capistrano doesn't seem to discern between failed and successful deployments for cleanup and rollback at all, which, to be honest, seems a bit strange to me for a deployment tool.

[mattbrictson]

Hi @tminich, thanks for the report and apologies for the lack of response.

capistrano doesn't seem to discern between failed and successful deployments for cleanup and rollback at all

That's correct. Capistrano does not have any way to mark a deployment as succeeded or failed. This is why keep_releases does not distinguish which releases get deleted.

I think this is a long-standing design decision and predates my time with Capistrano.

However, I definitely understand that this behavior could be surprising. Perhaps it should be documented somewhere? I would be open to accepting a PR that updates the keep_releases section of the documentation, for example.

capistrano/docs/documentation/getting-started/configuration/index.markdown

Lines 107 to 110 in efdc67b

	* `:keep_releases`
	* **default:** `5`
	* The last `n` releases are kept for possible rollbacks.
	* The cleanup task detects outdated release folders and removes them if needed.