Upgrading while there is pending publication requests causes havoc in a cluster.

If you upgrade a repository while there was pending publication request(s) in the DB, at restart we experience several issues:

• the Quartz scheduler will enter into a loop because it fails at deserializing the saved scheduled publications.

• the MySQL binlogs [1] are filled with attempts at setting a JCR lock on the quartz trigger node.

[1] insert into REPOSITORY_JOURNAL (REVISION_ID, JOURNAL_ID, PRODUCER_ID, REVISION_DATA) values (6135077,'cms-bnl2pa59','JR',_binary'\0\0default\0D\0\0
=¤zß\0NLpLQÆé1E©»ýÔ{%ı╩j\0P
I\0\0\0\0\0\0
jcr:lockOwner\0P
I\0\0\0\0\0\0jcr:lockIsDeep\0EI\0\0\0\0\0\0a/content/documents/brummen-voorst/agenda/robert-vuijsje/hippo:request/hipposched:triggers/defaultI \0\0
jcr:lockOwner\0\0\0\0\0\0hipposched:trigger\0\0\0
\0\0mix:lockable\0\0system\0EI\0\0\0\0\0\0a/content/documents/brummen-voorst/agenda/robert-vuijsje/hippo:request/hipposched:triggers/defaultI \0\0jcr:lockIsDeep\0\0\0\0\0\0hipposched:trigger\0\0\0
\0\0mix:lockable\0\0system\0\0')
/!/;