Sync now skips most steps if the feed metadata has not changed.

[mhrivnak]

closes #2

https://pulp.plan.io/issues/2

[mhrivnak]

The idea here is that if a previous sync has the skip_list set to skip a type, and then the user removes that from the skip list and does a sync, we don't want to skip the sync just because the metadata didn't change.

So if anything was removed from the previous skip list, we will do a full sync.

[beav]

That makes sense, thanks.

[beav]

assigning back to @mhrivnak per IRC convo

[beav]

@mhrivnak Would it make more sense to check timestamp from the primary data instead of revision? That might be what we are looking for. This is also what yum uses when examining metadata to determine if something is new.

for example:

<?xml version="1.0" encoding="UTF-8"?>
<repomd xmlns="http://linux.duke.edu/metadata/repo" xmlns:rpm="http://linux.duke.edu/metadata/rpm">
 <revision>1429551055</revision>
<data type="filelists">
  <checksum type="sha256">5cc15042c277bfbf8b04e8b1eadc9cf998016529663d504e93a79e0bff0246fa</checksum>
  <open-checksum type="sha256">df79c6949d62ecaff4f53bd4758b02b445f7eddd3a5b92eb8e6bdccfcd6203dd</open-checksum>
  <location href="repodata/5cc15042c277bfbf8b04e8b1eadc9cf998016529663d504e93a79e0bff0246fa-filelists.xml.gz"/>
  <timestamp>1429551057</timestamp>
  <size>312</size>
  <open-size>522</open-size>
</data>

revision is the epoch time but can be freeform text passed in by the user to createrepo. However, the timestamp might be a better candidate for what we are doing.

Another thought is that we could just save a hash of the repomd.xml itself and save it in the scratch pad 🐨. If the hash doesn't change, it would be OK to skip the other steps. This would free us from having to worry about specifics of repomd.xml creation.

[mhrivnak]

Good thoughts. I didn't go with the timestamp on primary.xml because it only covers RPMs. In a case where there are drpms, errata, groups, or any other custom metadata files, we need to know if any of those changed either.

revision seems like the "right" field for this. It is the part of the schema that is clearly meant to express whether or not the content has changed in a meaningful way (although docs would be nice to confirm the intent). I agree that using the file's checksum to detect change would work and could even be more reliable, but it removes the opportunity for someone to use the revision field appropriately. For example, someone could generate repo metadata without sqlite files, and then add them later, keeping the same revision.

In any case, I can't find an example of a yum repo where the revision is not an integer. I think we should give that a try, and if we find a use case where someone is committed to doing something different, we can consider fall-back strategies, like using the checksum.

We could additionally use cache-related HTTP headers to determine when this file, or others, hasn't changed, but that would be gravy.

[beav]

👍 sounds good, I am OK with using revision.

I don't remember if we need more changes or not; just assign back to me when it's time to re-review.

[mhrivnak]

Great. I am adding the improvement that it will default to full-sync if the revision can't be parsed as an int.

[beav]

I like the "if zero then always sync" approach. LGTM