CrashSimulator 3.0 is GA: prove your Oracle database comes back

Your backup job reported success. Your Data Guard dashboard is green. Your DR runbook was written two years ago and has never been executed end to end.

None of that is evidence. It is paperwork about evidence.

CrashSimulator 3.0 is generally available under Apache 2.0. It breaks an Oracle database on purpose, in a lab, drives the recovery, and keeps the receipts — logs, manifests, audit records and a report you can hand to an auditor without editing.

What you get

One shell script on the database host. No agent, no server, no repository database, no web console. You run it from the command line or drive it through the guided workflow menu, and it works on a standalone target.

123 scenarios — 103 database-host, application access-path and platform/readiness drills, plus 20 Autonomous Database cloud-service scenarios:

• Core — instance, datafile, redo and control-file failure (20)
• Multitenant PDB drills (17)
• Data Guard — FSFO, transport and apply lag, standby redo logs (10)
• APEX/ORDS application access-path drills (10)
• Backup (6) · OCI Base DB (5) · RAC (4) · Exadata (4) · GoldenGate (4) · Config (4) · ASM (3) · Corruption (3) · Services (3) · GI (2) · Compliance (2), plus Security, Network, Logical, Recovery and Lifecycle drills
• A separate Autonomous Database coverage model (ADB01–ADB20): logical and user-error recovery, clone/PITR readiness, wallet and private-endpoint connectivity, Autonomous Data Guard, IAM, Object Storage and resource limits

Every run produces a Resilience Score out of 100, split into domain scores for backup, RAC/local HA, security, DR, recoverability, MAA alignment, scenario coverage and application continuity. That is the number you take into a management or audit conversation, and it is derived from drills that actually ran.

This release is about honesty, not coverage

3.0 adds very little to the catalogue. It fixes something more important: every place where the tool knew one thing and reported another.

• Scenarios with no automated recovery say so before the fault, not after it.
• A destructive plan whose files are not backed up is refused up front.
• Validation-only drills no longer offer to recover something they never broke.
• Total redo loss decides from live v$log state instead of assuming.
• ASM write privilege is probed before anything goes offline — a missing grant is a clean refusal that changed nothing, not a half-executed drill.
• Credentials never appear on a sqlplus command line.
• Un-recovered drills block new destructive ones, and --reconcile-drills closes out the ones the live database shows are already recovered.

Every competitor claims coverage. This release claims accuracy about its own limits. A tool that breaks databases has to be exact about what it did.

Five foundations

Security. Availability. Recoverability. Integrity. Evidence.

Safety

Lab, development, training and resilience-test environments. Do not run destructive scenarios in production. The engine plans in dry-run first and refuses what it cannot finish.

Run your first drill

github.com/fmunozalvarez/crashsimulator

Download crashsimulator-v3.0.3-runtime.zip, verify the published SHA-256, unzip on the database host:

./CrashSimulatorV2.sh –discover ./CrashSimulatorV2.sh –menu

Oracle Database 12c and later. Validated against live Oracle 19c and Oracle AI Database 26ai RAC/ASM labs — CrashSimulator project validation, not Oracle certification.