Separate recovery from defect removal
Recreating the pool makes new resources available and permits an operation, but the same faulty sequence retains three connections again. Increasing capacity to five keeps two free after the first three failures; after the fifth, it also exhausts. Temporary relief can help complete a fictional batch when authorized and proportionate to risk. Record applicability, costs, impact, evidence to preserve and stopping criteria. Do not declare the cause removed because the latest batch finished. The decision can retain an owned correction or accept temporary risk through applicable governance. The sponsor provides priorities, but a request to close a record does not replace technical criteria and defined authority for accepting residual risk.
Exercise the trigger and invariant
Corrected regression alternates 30 successes, 30 exceptions and 30 cancellations. After every cycle, three connections are free and none remains leased. Exception and cancellation do not disappear: the observed change is capacity recovery without changing task outcome. The concurrency group adds waiting and reuse; the closed-connection group checks that a positive count is insufficient. These tests connect the mechanism to acceptance conditions. Ninety exits are a workshop choice, not statistical proof or coverage of every failure. Keep code and driver versions, sequence, expected and observed results. When a patch passes only success, request regression of paths causing recurrence instead of more volume on the same path.
Move from the lab to the affected product
A small reproduction can clarify a hypothesis without proving the real product fails in the same way. For an application using WebSphere, JDBC or Oracle, confirm version, pool configuration, close and cancellation contracts, transactions and behavior when a dependency is lost. The local lab did not execute those technologies. Define an authorized exercise with representative load and paths, functional outcomes, resource accounting and exposure limits. Retain a stop criterion and recovery alternative while assessing workaround retirement. An incident-free period may coincide with blocked demand or an absent trigger; document exposure. If two services share a timeout but one retains resources and the other executes long queries, preserve distinct investigations until evidence supports a common mechanism.
Track distinct outcomes through acceptance
A review may identify omitted return, testing without cancellation and CPU-only alerting. These factors are related but need different outcomes: correct the mechanism, prevent regression and detect impact through suitable signals. A new dashboard may satisfy the third action without completing the others. For each action, record owner, deadline, evidence and acceptance condition; review residual risk and capacity with people able to decide. Update the knowledge base by version and retain workaround history when it ceases to apply. Prepare an English supplier review using facts, gaps and next tests. The guide includes a proposed workshop, but it has not been performed by a human group and has not received independent specialist review.
python3 - <<'PY'
import json
from pathlib import Path
e=json.loads(Path('content/labs/problem-runtime/evidence.json').read_text)
for name in ['recreationDoesNotFixLeak','largerPoolDelaysExhaustion','repeatedExitPaths']:
print(name, e['checks'][name])
PYRecreating or enlarging the pool permits temporary progress, but repeating the trigger consumes capacity again.
Common pitfalls
Confusing workaround with correction, ninety exits with complete coverage or a delivered dashboard with effective completion of every action.
Related topics: Investigation and evidence · Workarounds and correction acceptance
Accept demonstrated outcomes within the correct scope and retain investigation or formally tracked risk where evidence is missing.
Reference: OPS11-BP02 Perform post-incident analysis · Problem management practices 2026-09; ServiceNow Brazil examples with scoped plugins and properties