The repository manager is the queue manager component that publishes cluster object definitions from full repositories to partial members and keeps cluster catalogs aligned—covered conceptually in the repository manager tutorial. Repository manager problems are the failure mode: DEFINE completes locally but never reaches partners, REFRESH CLUSTER returns errors, error logs mention publish or repository failures, or SYSTEM.CLUSTER.REPOSITORY.QUEUE behaves abnormally on a full repository host. Beginners sometimes confuse these faults with generic channel problems; the distinction is that CHSTATUS may show RUNNING on an application cluster channel while the path to the full repository is still broken, starving the repository manager of updates. This troubleshooting guide maps symptoms to causes, log and DISPLAY evidence, repository queue health checks, role misconfiguration on REPOS and REPOSNL, interaction with stuck cluster channels, safe recovery ordering, and when to escalate beyond REFRESH CLUSTER.
When an administrator defines a cluster queue, the local queue manager accepts the object into its local repository and the repository manager formats a publication toward full repository queue managers. Full repositories merge the definition into the authoritative catalog on SYSTEM.CLUSTER.REPOSITORY.QUEUE and fan updates to partial subscribers. Partial repository managers apply updates to cache. Any break in that chain—no REPOS host reachable, publish inhibited, corrupt message, wrong cluster name—surfaces as manager problems. Application puts may still work locally on the defining box, masking the defect until remote routing is attempted.
| Symptom | Likely cause | Action |
|---|---|---|
| DEFINE not visible on any remote | No path to full repo | Fix CLUSSDR to REPOS host |
| Visible on one full repo only | Split or partial publish | Compare both full repos |
| REFRESH CLUSTER fails | Role or queue inhibit | DISPLAY QMGR REPOS, QSTATUS system queue |
| Intermittent sync | Flapping channel | Stabilize network, tune retry |
| Errors after restore | Old repository queue data | IBM recovery procedure |
123456DISPLAY QMGR CLUSTER CLUSNL REPOS REPOSNL DISPLAY QSTATUS('SYSTEM.CLUSTER.REPOSITORY.QUEUE') DISPLAY CHSTATUS(*) WHERE(CHLTYPE EQ CLUSSDR) * Error log on full repository host around DEFINE time * Compare CLUSQ on full repo immediately after DEFINE: DISPLAY CLUSQ('NEW.QUEUE') CLUSTER('SALES')
Save queue manager error log excerpts with AMQ messages referencing repository or cluster publish—search IBM Knowledge Center for your exact message number. Note whether DEFINE was issued on a partial member without connectivity to full repository at that second. Timestamp correlation with channel STOP events narrows root cause quickly.
On full repository hosts, this system queue holds catalog messages. CURDEPTH should move during administrative churn but not grow unbounded without explanation. IPPROCS or OPPROCS inhibited, queue full, or media errors block repository manager writes. DISPLAY QSTATUS shows inhibit flags and depth. Do not DELETE messages from this queue without IBM guidance—manual deletion destroys catalog integrity. If damage is suspected after storage failure, engage support with backup and recovery options for the queue manager.
A queue manager with CLUSTER but no working path to any full repository cannot publish outward. A host mistakenly marked REPOS without planning becomes an unintended full repository—other members may publish to it while operations thought only two designated hosts existed. DISPLAY QMGR on every cluster member and maintain a spreadsheet of REPOS and REPOSNL roles. Decommission removes CLUSTER, CLUSNL, REPOSNL, and cluster channels in planned order—see partial repositories tutorial for leave procedure.
The repository manager is the librarian who mails updated card catalog cards to every branch. Problems mean the librarian's outbound mail is stuck in the office—branches keep filing books with outdated cards even when local staff work hard.
The teacher who tells every classroom the new rules cannot make phone calls today—so some classrooms never hear the new recess time.
List log and DISPLAY evidence to collect before REFRESH CLUSTER.
Explain why RUNNING app channel does not prove repository sync.
When is manual repository queue intervention never acceptable?
1. Repository manager runs:
2. Catalog messages on full repo often on:
3. Publish failure may show as:
4. First connectivity check: