# Oracle Dba Exadata Engineered Systems

> USE THIS SKILL WHENEVER the user mentions Exadata, Exadata Database Machine, Engineered Systems, storage cells / cellcli / dcli / griddisk / celldisk / physicaldisk, Smart Scan / cell offload / storage index, Exadata Smart Flash Cache / flash log / CELL_FLASH_CACHE, IORM / IO Resource Manager / IORMPLAN, exachk / AHF / TFA on Exadata, patchmgr / Exadata patching (cells, RoCE/InfiniBand switches, compute/db nodes, image/RU), Exadata expansion / rack scale-out / add cells / ASM rebalance on Exadata, Exadata Cloud@Customer storage cells, or Exadata-side ASM disk-group / failgroup work — EVEN IF they do not name the specific task. Covers (T1) exachk/AHF health + in-DB cell visibility, cell-status LIST, IORM/DBRM read, Smart Scan / flash / expansion diagnostics; (T2, --dry-run default) patchmgr PRECHECK, in-DB DBRM plan activation, and the CELL_FLASH_CACHE segment hint; and (T3, runbook-only) cell/switch/compute patch APPLY, storage-cell management, cell IORMPLAN configuration, cell flash-cache tuning, and rack ex

- Skill: `shreyas70773/oracle-dba-exadata-engineered-systems` (Agent Skill, multi-file: 23 files)
- Install (CLI): `npx skillmds@latest add shreyas70773/oracle-dba-exadata-engineered-systems`
- Raw SKILL.md: https://api.skillmd.com/api/skills/shreyas70773/oracle-dba-exadata-engineered-systems/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: Shreyas70773 (https://skillmd.com/u/shreyas70773)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/shreyas70773/oracle-dba-exadata-engineered-systems

---

# Oracle DBA - Exadata / Engineered Systems

Health, storage-cell visibility, IORM/Smart-Scan/flash diagnostics, patch readiness, and
the plan-only Tier-3 runbooks for Oracle Exadata Database Machine (19c and 23ai; EE;
single-instance and Data Guard; on-prem). Two mechanisms by design:

- **In-DB** work (offload stats, ASM headroom, DBRM plans, CELL_FLASH_CACHE attributes)
  runs through the **SQLcl MCP** saved connection `dba_ai_conn` → `sql/*.sql` ("sqlcl-mcp").
- **Host** work (exachk/AHF, `cellcli`/`dcli`, `patchmgr`) **cannot** go through SQLcl MCP
  (restrict level 4) → `.sh`/`.ps1` ("script", to later front a guarded oracle-dba-ops MCP).

Everything self-executing is **T1 (observe)** or **T2 (reversible, dry-run-default)**. Every
cell/switch/compute **patch apply**, **storage-cell mutation**, **cell IORMPLAN config**,
**cell flash-cache tuning**, and **rack expansion** is **T3** and lives only in
`references/runbooks.md` — it prints commands for a human and never self-executes.

## Scope & risk map

| Task | Level | Tier | Mechanism (script / sql / runbook path) | Idempotent? |
|------|-------|------|------------------------------------------|-------------|
| Exachk / health check execution (exachk/AHF + in-DB summary) | L2 | T1 | `scripts/run_exachk.sh` / `.ps1` ("script") + `sql/exadata_health_summary.sql` ("sqlcl-mcp") | yes — read-only; run-in-progress lock no-ops; failing best practice = finding |
| In-DB Exadata health & cell-visibility summary | L2 | T1 | `sql/exadata_health_summary.sql` ("sqlcl-mcp") | yes — read-only |
| Storage-cell status (read-only management visibility) | L3 | T1 | `scripts/cell_status.sh` / `.ps1` ("script", cellcli/dcli LIST) | yes — read-only; off-cluster no-ops |
| IORM visibility (cell IORMPLAN + in-DB DBRM) | L3 | T1 | `scripts/iorm_status.sh` / `.ps1` ("script") + `sql/iorm_status.sql` ("sqlcl-mcp") | yes — read-only |
| Smart Scan / flash-cache diagnostics | L3 | T1 | `sql/smartscan_diagnostics.sql` ("sqlcl-mcp") | yes — read-only |
| Expansion / capacity readiness (pre-flight facts) | L3 | T1 | `sql/expansion_readiness.sql` ("sqlcl-mcp") | yes — read-only |
| Exadata patch READINESS PRECHECK (cells/switch/compute) | L2 | T2 | `scripts/patch_precheck.sh` / `.ps1` ("script", patchmgr `-precheck` only) | yes — precheck is read-only/repeatable; refuses to apply |
| In-DB DBRM plan activation (in-DB half of IORM) | L3 | T2 | `scripts/dbrm_plan_apply.sh` / `.ps1` ("script") + `sql/dbrm_plan_apply.sql` ("sqlcl-mcp") | yes — already-active → noop; plan-not-found → blocked |
| Smart Flash Cache segment hint (CELL_FLASH_CACHE KEEP/DEFAULT/NONE) | L3 | T2 | `scripts/flashcache_hint.sh` / `.ps1` ("script") + `sql/flashcache_hint_apply.sql` ("sqlcl-mcp") | yes — attribute already set → noop; metadata-only |
| Exadata patching APPLY (cells / IB-RoCE switches / compute + GI/RDBMS) | L2/L3 | **T3** | `references/runbooks.md#exadata-patching` | no — reboots components; rolling per cell/node |
| Storage-cell management (griddisk/celldisk/disk/firmware) | L3 | **T3** | `references/runbooks.md#cell-management` | no — DROP GRIDDISK is data-destructive |
| IORM configuration (cell IORMPLAN + DBRM plan DEFINITION) | L3 | **T3** | `references/runbooks.md#iorm-config` | no — reshapes I/O priorities estate-wide |
| Smart Scan / flash-cache tuning (cell-side CREATE/ALTER FLASHCACHE) | L3 | **T3** | `references/runbooks.md#flashcache-tuning` | no — drops cached/dirty data (flush first) |
| Exadata expansion / rack scale-out (add cells/nodes + ASM rebalance) | L3 | **T3** | `references/runbooks.md#expansion` | partial — readiness repeatable; `compatible.*` raise is one-way |

## Preconditions

- **SQLcl MCP connection `dba_ai_conn`** is reachable and wallet-backed. All in-DB
  prechecks, postchecks, idempotency probes, and `sql/*.sql` run through it via
  `connect_db` / `Connect-Db`. Never SYS/SYSTEM — except inside a T3 runbook that
  explicitly states `AS SYSDBA` / `AS SYSASM` (e.g. ASM `ALTER DISKGROUP`), supplied
  interactively by the operator.
- **Secrets resolve from the Oracle Wallet / external password store. No plaintext
  anywhere.** Host tools (`patchmgr`, `cellcli`, `dcli`, `exadcli`, `ahfctl`, `asmcmd`,
  `srvctl`) use OS / key-based SSH set up out of band — never a password on argv. The
  shared helpers reject credential-looking arguments with `ERR_SECRET` (8).
- **Helpers are sourced from `_common`** (`scripts/lib.sh` / `lib.ps1`): `connect_db`,
  `log_event`, `emit_metric`, `require_dry_run`, `guard_blocked_dry_run`,
  `require_approval_token`, `precheck`, `postcheck`, `print_banner`. Not reinvented.
- **Host reachability**: cell/IORM/patch scripts need a node from which `cellcli`/`dcli`/
  `patchmgr` can reach the cells/switches/db-nodes via SSH keys, with the group files
  (`cell_group`, `ibswitch_group`, `dbs_group`) present. Off-cluster, the read-only
  scripts log `noop` and fall back to the in-DB summary.
- **Logging**: structured `key=value` lines to the canonical paths
  `/var/log/oracle-dba/exadata/exadata.log` (Linux) /
  `C:/ProgramData/oracle-dba/logs/exadata/exadata.log` (Windows); skill segment `exadata`.
- **T2 dry-run default**: every T2 script previews and changes nothing unless given
  `--execute`. `patch_precheck` additionally REFUSES to apply patches even with `--execute`
  (it only runs `patchmgr -precheck`); the apply is T3.

## Procedures

Scripts ship as behaviour-equivalent `.sh` (Linux) / `.ps1` (Windows) pairs. T2 scripts
default to `--dry-run`; pass `--execute` to act. See each script's `--help`.

### Exadata health check (exachk / AHF) — T1
- **Trigger**: weekly best-practice health check; before/after any cell/patch/expansion work.
- **Precheck**: `dba_ai_conn` reachable (for the in-DB summary half).
- **Action**: `scripts/run_exachk.sh --silent` / `.ps1 -Silent` ("script") + `sql/exadata_health_summary.sql` ("sqlcl-mcp"). `--skip-indb` for host-only.
- **Postcheck**: n/a — emits metrics (cell_count, asm headroom, rebalance) + report path; a failing best practice is a finding, not a failure.
- **Rollback note**: none — read-only.

### Storage-cell status — T1
- **Trigger**: daily; investigating a non-NORMAL disk or a critical cell alert.
- **Precheck**: none (read-only); auto-detects the cell group / runner.
- **Action**: `scripts/cell_status.sh [--detail]` / `.ps1 [-Detail]` ("script", cellcli/dcli LIST only).
- **Postcheck**: n/a — surfaces unexamined critical alerts + non-NORMAL griddisks as findings.
- **Rollback note**: none — read-only. Cell mutation is T3 `#cell-management`.

### IORM visibility — T1
- **Trigger**: confirm the active IORM/DBRM picture across cells + DB.
- **Precheck**: `dba_ai_conn` reachable (in-DB half).
- **Action**: `scripts/iorm_status.sh` / `.ps1` ("script", cell `LIST IORMPLAN`) + `sql/iorm_status.sql` ("sqlcl-mcp").
- **Postcheck**: n/a — read-only.
- **Rollback note**: none. Cell IORMPLAN config is T3 `#iorm-config`; activating an existing DBRM plan is the T2 below.

### Smart Scan / flash & expansion diagnostics — T1
- **Trigger**: "why am I (not) getting Smart Scan/flash benefit?" / pre-expansion fact-finding.
- **Precheck**: `dba_ai_conn` reachable.
- **Action**: `sql/smartscan_diagnostics.sql` / `sql/expansion_readiness.sql` ("sqlcl-mcp").
- **Postcheck**: n/a — read-only; emits offload/headroom KEYVAL metrics.
- **Rollback note**: none — read-only.

### Exadata patch readiness PRECHECK — T2
- **Trigger**: before a quarterly image/RU; validate a component CAN be patched.
- **Precheck**: component valid; on `--execute`, patchmgr present + patch dir / target list exist.
- **Action**: `scripts/patch_precheck.sh --component cells|switch|compute [--patch-dir DIR] [--target-list FILE] --execute` / `.ps1` ("script"). Dry-run prints the exact `patchmgr -precheck` command; `--execute` runs the precheck (still non-destructive).
- **Postcheck**: precheck output reports READY (no FAILED/ERROR).
- **Rollback note**: none — precheck never applies. The APPLY is T3 `#exadata-patching`.

### In-DB DBRM plan activation — T2
- **Trigger**: switch the active in-DB resource plan (the in-DB half of IORM on Exadata).
- **Precheck**: `dba_ai_conn` reachable; the requested plan must already EXIST (else blocked).
- **Action**: `scripts/dbrm_plan_apply.sh --plan NAME --execute` / `.ps1 -Plan NAME --execute` ("script") + `sql/dbrm_plan_apply.sql` ("sqlcl-mcp"). Dry-run reports the current plan only.
- **Postcheck**: requested plan is the active top plan (or already-active no-op).
- **Rollback note**: reversible — re-activate the previous plan (printed). Creating/altering a plan DEFINITION is T3 `#iorm-config`.

### Smart Flash Cache segment hint — T2
- **Trigger**: pin a hot small table/index in flash (KEEP) or release it (DEFAULT/NONE).
- **Precheck**: `dba_ai_conn` reachable; owner/table valid; mode in KEEP|DEFAULT|NONE.
- **Action**: `scripts/flashcache_hint.sh --owner O --table T --mode KEEP --execute` / `.ps1 -Owner O -Table T -Mode KEEP --execute` ("script") + `sql/flashcache_hint_apply.sql` ("sqlcl-mcp").
- **Postcheck**: `CELL_FLASH_CACHE` attribute equals the requested mode.
- **Rollback note**: reversible — set `--mode DEFAULT`. Cell-side flash sizing/mode is T3 `#flashcache-tuning`.

## Tier-3 runbooks

All **destructive Exadata work** is Tier 3 and lives in **`references/runbooks.md`**:
`#exadata-patching` (cell/switch/compute apply + GI/RDBMS/datapatch),
`#cell-management` (griddisk/celldisk/disk/firmware), `#iorm-config` (cell IORMPLAN + DBRM
plan DDL), `#flashcache-tuning` (cell flash cache CREATE/ALTER/DROP), and `#expansion`
(add cells/nodes + ASM rebalance).

These runbooks **PRINT exact cellcli/patchmgr/ASM/SQL commands for a human operator and
NEVER self-execute.** They are gated by `require_approval_token` / `Require-ApprovalToken`:
an automated agent's only sanctioned action is to print the relevant section and stop with
`ERR_APPROVAL` (6) until a human supplies a real change ticket/token (`--token <TICKET>` or
`ODB_APPROVAL_TOKEN`; placeholders are rejected). The T2 scripts deliberately refuse the
destructive paths (`patch_precheck` never applies; the wrappers never define plans or touch
cell hardware) and point to the runbook instead. SYSDBA / SYSASM is a T3-only,
interactively-supplied exception. When in doubt, runbook.

See `references/version-notes.md` for 19c-vs-23ai (and Exadata-image) behavioural deltas,
and `cron/crontab.d` / `cron/task-scheduler.xml` for the schedulable T1/T2 jobs.