Bug 871886

Description of problem:

Manifest imports can fail when content IDs have changed upstream.

This is very similar to bug #807468 where we were relying on content labels, which turned out to change sometimes, so we switched to content ID as the definitive way to check if the content already exists on the server.

How reproducible:

Difficult, need a manifest with an old product ID and a content label, then a newer manifest with the same label but a different ID.

Saw this with:


11:17am <dgoodwin> thomasmckay: ok this looks to be because rel-eng changed the content id for rhel-5-server-rhev-agent-rpms
11:17am <dgoodwin> from 1105 which you have in your db from an old manifest
11:17am <dgoodwin> to 1687


The error will surface like so in catalina.out:

Oct 31 10:27:59 [http-8443-1] DEBUG org.candlepin.sync.Importer - Import product: RH0100931.json
Oct 31 10:27:59 [http-8443-1] DEBUG org.candlepin.sync.Importer - Import product: RH0197181.json
Oct 31 10:27:59 [http-8443-1] DEBUG org.candlepin.sync.Importer - Import product: RH0149709.json
Oct 31 10:27:59 [http-8443-1] DEBUG org.candlepin.sync.Importer - Import product: 180.json
Oct 31 10:27:59 [http-8443-1] DEBUG org.candlepin.sync.Importer - Import product: RH0197181F3.json
Oct 31 10:27:59 [http-8443-1] DEBUG org.candlepin.sync.Importer - Import product: 69.json
Oct 31 10:27:59 [http-8443-1] DEBUG org.candlepin.sync.Importer - Import product: RH0101594.json
Oct 31 10:27:59 [http-8443-1] WARN  org.hibernate.util.JDBCExceptionReporter - SQL Error: 0, SQLState: 23505
Oct 31 10:27:59 [http-8443-1] ERROR org.hibernate.util.JDBCExceptionReporter - Batch entry 0 insert into cp_content (created, updated, contentUrl, gpgUrl, label, metadataExpire, name, releaseVer, requiredTags, type, vendor, id) values ('2012-10-31 10:27:59.424000 -04:00:00', '2012-10-31 10:27:59.424000 -04:00:00', '/content/dist/rhel/server/5/$releasever/$basearch/rhev-agent/3/os', 'file:///etc/pki/rpm-gpg/RPM-GPG-KEY-redhat-release', 'rhel-5-server-rhev-agent-rpms', '86400', 'Red Hat Enterprise Virtualization Agents for RHEL 5 Server (RPMs)', NULL, 'rhel-5-server', 'yum', 'Red Hat', '1687') was aborted.  Call getNextException to see the cause.
Oct 31 10:27:59 [http-8443-1] WARN  org.hibernate.util.JDBCExceptionReporter - SQL Error: 0, SQLState: 23505
Oct 31 10:27:59 [http-8443-1] ERROR org.hibernate.util.JDBCExceptionReporter - ERROR: duplicate key value violates unique constraint "cp_content_label_key"
Oct 31 10:27:59 [http-8443-1] ERROR org.hibernate.event.def.AbstractFlushingEventListener - Could not synchronize database state with session
org.hibernate.exception.ConstraintViolationException: Could not execute JDBC batch update




We are unclear how to fix at this time. If rel-eng can prevent the content IDs from ever changing we will be ok.

If not we have to figure out what to do in Candlepin. Some possibilities:

1. Remove the label uniqueness contraint, let the content go in with the new ID. Products will be pointing at that new ID anyhow. Content could get orphaned but that's easy to detect and clean up as well.

2. Treat content like product attributes, all products get their own copy of content. Lot of things to consider here like upgrade scenarios, promotion of content to environments (which is by ID and expected to trigger for all products using that content, there could be many)


Need to proceed carefully in any case.