Oracle-DBA-Study

1 · RAC node eviction during batch

A 2-node RAC cluster evicts node 2 every night during batch, with gc cr block busy waits spiking first. Is this a clusterware bug, and what single check separates network cause from load cause?

Answer: not a bug until proven — check interconnect heartbeat history + OS network drops against eviction timestamps in CRS logs. Flaky switch = heartbeat loss with low load; real overload = waits track batch CPU/IO. Trap: patching clusterware before ruling out the interconnect.

2 · DataGuard two hours behind before close

Standby is 2h behind at month-end. V$DATAGUARD_STATS shows transport lag 0, apply lag 115 min, MRP running. Do you touch the network, and what is the fix order?

Answer: do not touch the network — transport is keeping up, apply is not. Fix order: standby disk/IO rate → missing index slowing SQL apply → parallel apply/MRP tuning → then backlog drain + lag alerting. Trap: failing over to a standby that has not caught up.

3 · GoldenGate replicat lag after batch

Replicat lag grows on the reporting target after a batch load; extract and pump show no lag. Add parallelism immediately?

Answer: not yet — check the target apply plan first (missing index on the target is the classic cause). Fix: index or parallelize replicat, drain backlog, add per-process lag monitors. Trap: tuning the source when only replicat lags.

Daily say-aloud reps

• "Understand. Prepare. Identify. Support. Data first, trap ruled out, fix that stays fixed."
• "RAC is not HA without fencing and quorum — interconnect first, clusterware second."
• "Switchover is planned and lossless; failover is unplanned and may lose data."
• "Every bi-directional table gets a conflict rule before go-live, or replication loops."
• "Top wait event first, top SQL second, one hypothesis, then test."
• "An untested backup is not a backup — restore proves it."
• "Customer impact first, root cause second, stay-fixed third."
• "Compliance is evidence — every privileged action logged and reviewable."
• "My gap is hands-on AWS-native depth versus Oracle internals — closing it with DMS, RDS, and CloudWatch labs in this 10-day plan."

Drill 1 · AWR read: name the bottleneck

Top-5 timed events: gc cr block busy 38%, db file sequential read 22%, CPU 15%. Top SQL by that wait is an order-lookup joining on an unindexed status column. Narrate: top line first, the one fix with biggest payoff, and what you re-measure.

Bug → risk → fix → test. Fix: composite index on the filter+join columns (equality first, range last), then check interconnect if GCS waits persist. Re-measure: AWR top-event share + SQL elapsed time for that sql_id vs baseline. Trap: tuning the I/O the AWR ranks second.

Drill 2 · Buggy RMAN backup shell

#!/bin/sh
rman target / <<EOF
BACKUP DATABASE;
DELETE NOPROMPT OBSOLETE;
EOF
echo "backup done"

Name every bug, its production risk, the fix, and one test case each — before editing.

Bugs: no archivelog backup (unrecoverable gap) · no controlfile/spfile autobackup check · no VALIDATE or restore test (untested backup) · no retention/RPO match · silent failure (exit code unchecked, no alerting) · credentials/log exposure risk. Fix: archivelogs + autobackup on + restore-to-test proof + set -e with alerting. Test: restore to auxiliary + PITR rehearsal.

Drill 3 · GoldenGate lag check one-liner

Replicat lag is climbing. Write the GGSCI check sequence from memory, state which process the lag belongs to, and the metric that proves recovery.

INFO ALL → LAG EXTRACT * → LAG PUMP * → LAG REPLICAT * + trail-file growth + checkpoint age. Prove with per-process lag back to baseline for a full batch cycle. Trap: restarting replicat before reading the checkpoint.