Generic tags and named snapshots : is there some difference?

[paul.zakharov]

This is some kind of continuation of questions :

Tags and named transactions : clarification is needed

I’d like to have some advice about small problem. We have two publishers connected with gateway installed on Stratum0. Transactions on our system is triggered by some CI system. So in general it’s something that we don’t control manually. Recently we add the tag option “-a” to our publish command. But the main (and almost the only one) reason for that was to be able to detect which process/commit trigger the CI (and so trigger the publish). Finally we are re-reading docs and found that tag …

and

Publish -m = tag description without tag?

This is the new part of Tags and named transactions : clarification is needed topic (the original post is closed now). For my question number #2 : Are there some options for giving a name (or comments) to transaction but without using tag ? From docs it seems that no, but I prefer to ask. I’d like to know if I could submit a feature request to have such option ? We found that publish command has “-m” option : [-m tag description] It seems that it does the job for, we could create comment…

Some points are still unclear for us.

1. if we don’t use “-a” option our publish command still create some generic tags, like generic-[date and time].
   I suppose it’s because we have CVMFS_AUTO_TAG=true is our repo settings.

2. But also we have CVMFS_AUTO_TAG_TIMESPAN=“” in the settings.
   Does it means that our generic tags are never removed ?

3. If the answer to #2 is yes.
   Does our transactions with generic tags are never impacted by garbage collector ?
   So globally it means that our garbage collector remove nothing. Am I right ?

4. Final question (for the moment).
   If it’s not the case, and generic tags are impacted by garbage collector.
   Is there some way to distinct “generic” and “named” snapshot except using some name pattern ?

Thank you very much for helping.

[dwd]

The answer to numbers 1, 2, and 3 is yes. I really think that CVMFS_AUTO_TAG_TIMESPAN should default to something like 4 weeks ago on newly created repositories. I’ll make an issue about that.

[paul.zakharov]

Thank you very much for your reply. It’s very precious for us.

I wondering, if now we set timestamp for something, for example
CVMFS_AUTO_TAG_TIMESPAN=“3 months ago”.

Which process will remove generic tags ? Should we start it manually or it will be automatic ?

We actually have more than 1000 tags in our repository.
Could setting TIMESPAN provoke some downtime ?

[dwd]

The next publish command should remove all the tags older than that. It won’t take very long and won’t provoke any downtime.

[dwd]

On the other hand, the next garbage collection that is done after that is likely to take longer while it cleans out deleted files no longer referenced by tags. How much longer depends on how big the repository is and how much is deleted, but it is potentially significant. Garbage collections do exclude other transactions both on the stratum 0 and stratum 1 while they’re running.