All topics
60 / 60
Adminintermediate

Monitoring and Health-Check Strategy for EBS Environments

Explain a comprehensive proactive monitoring approach spanning application, middleware, and database tiers in EBS

Mature EBS operations means finding problems before users report them. Three tiers, each with its own signals.

Comprehensive EBS monitoring is like a hospital's multi-system patient monitor rather than just a single heart-rate sensor — you need visibility into several distinct vital signs simultaneously (concurrent manager 'pulse', workflow 'circulation', middle-tier 'breathing', database 'blood pressure') because a problem in any one system can be the actual root cause even if another system's readings look fine.

Key Concepts

1
Application tier.
    concurrent managers   queue depth, requests pending too long,
                          actual against target processes
    workflow backlog      errored or deferred activities accumulating
    notification mailer   is it running
2
The mailer is the classic silent failure. It stops, and nothing errors. Workflows continue, notifications queue in the Worklist, and approvals stop happening because nobody was emailed.
3
Days later somebody asks why their requisition was never approved. Monitoring it is the only way this is found early.
4
The workflow backlog is similar.
    SELECT activity_status, COUNT(*)
      FROM wf_item_activity_statuses
     WHERE activity_status IN ('ERROR','DEFERRED')
     GROUP BY activity_status;
5
A growing deferred count means the Workflow Background Process is not running. A growing error count means something is failing repeatedly and nobody is looking at it.
6
Middle tier. JVM heap for the OA Framework processes, response times, and error rates.
7
Self-service page slowness is usually memory pressure — a JVM approaching its limit spends its time collecting garbage, so performance degrades before anything fails. Heap trending is what catches it.
8
Database tier. Tablespace growth, invalid objects, long-running sessions, locks and blocking sessions, and the alert log.
9
Invalid objects after a patch deserve their own check. A patch that completed with invalid packages leaves functionality that fails only when exercised, which may be at month-end.
10
What to alert on, as opposed to what to collect.
    the mailer is down
    concurrent requests pending longer than N minutes
    workflow errors above a threshold
    tablespace above 85 per cent
    a session blocking others for more than N minutes
11
And the principle behind the list. Alert on things that are silently wrong. A system that is plainly down generates its own alerts — the users. The value of monitoring is in the failures that produce no symptom until far too late, and every item above is one of those.
12
What the interviewer is probing.1. "What would you monitor daily?" Probing: the operational discipline. Stalls: "Is it up?" Moves up: concurrent manager status and queue depth, long-running requests, workflow errors, tablespace growth, and failed interfaces.
13
2. "Why is queue depth more useful than uptime?" Probing: the insight. Stalls: "It is not." Moves up: the system can be up while nothing is processing, and users experience that as an outage.
14
3. "What is the value of trending rather than thresholds?" Probing: the maturity point. Stalls: "None." Moves up: a job that normally runs 10 minutes taking 40 is a problem long before any threshold fires.
15
4. "What should a health check produce?" Probing: the process. Stalls: "A report." *Moves up:* an owned action list — a report nobody acts on is a cost, not a control.
Use cases:monitoringhealth-checkconcurrent-manageroemoperations