Files
rb/od-db-backup.md
T
igodwin adf6aa7658 Add od-backup-verify runbook: isolated Hyper-V test-restore + health checklist
Verify an od-db-backup cold backup by restoring it into a throwaway,
network-isolated Hyper-V VM (revert to checkpoint after), never over live
prod. Covers matching versions from MANIFEST, whole-datadir restore for
InnoDB, re-pointing the image path, and a pass/fail health checklist
(connects, Help>About version, Database Maintenance Check, recent data,
images open, no missing tables, optional mysqlcheck). Cites Open Dental
Backups / Manual Backups / Database Maintenance. Cross-linked with
od-db-backup.md; README updated.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-13 01:01:55 -07:00

147 lines
11 KiB
Markdown

# Runbook: Open Dental — Rock-solid database + images backup (Windows)
**Applies to:** The Windows machine hosting the Open Dental MySQL/MariaDB database (the "server").
**Goal:** A fully consistent, restorable backup of both halves of an Open Dental practice — the **database** and the **A-to-Z images folder** — following Open Dental and MySQL/MariaDB best practice, including cleanly **stopping the database during the copy and restarting it after**.
### What a complete Open Dental backup is (and isn't)
Per Open Dental's own docs, a full backup is **two things — the database and the A-to-Z images folder** — and both must be captured together. A database backup **without** the images folder (or vice-versa) is not a usable restore.
| Component | Default location | Back up? | Holds |
|---|---|---|---|
| MySQL/MariaDB **data directory** | `C:\mysql\data\` (contains the `opendental` DB) | **Yes — required** | All clinical/financial data |
| **A-to-Z / OpenDentImages** folder | `C:\OpenDentImages\` | **Yes — required** | Scanned docs, images, attachments |
| **`FreeDentalConfig.xml`** | Open Dental install dir | Nice-to-have | Connection / Choose-Database / AtoZ path config |
| Open Dental **program files** | `…\Program Files\Open Dental\` | **No — reinstall** | The application itself |
| Other office documents | (varies) | Per your policy | Non-Open-Dental files |
**There is no separate "application backup."** Open Dental does **not** back up the program itself — on restore you **reinstall the matching Open Dental version** (and matching MySQL/MariaDB) and point it at the restored data. So the two "recovery aids" worth grabbing are just: a copy of **`FreeDentalConfig.xml`** and a record of the **exact versions** to reinstall. The script captures both automatically into the backup folder (`FreeDentalConfig.xml` + `MANIFEST.txt`) — nothing extra to run.
**Placeholders:**
| Placeholder | Meaning |
|---|---|
| `<DB_SERVICE>` | Name of the MySQL/MariaDB Windows service (e.g. `MySQL`, `MySQL57`, `MariaDB`) |
| `<DATA_DIR>` | MySQL data directory — the folder **containing** the `opendental` subfolder (commonly `C:\mysql\data`) |
| `<IMAGES_DIR>` | A-to-Z images folder (commonly `C:\OpenDentImages`) |
| `<DEST>` | Backup destination — separate physical disk or UNC path, ideally replicated off-site |
| `<DB_USER>` / `<DB_PASSWORD>` | A MySQL account for `mysqldump` (supplemental method) — from the password manager, never committed |
---
## Why "cold copy" is the rock-solid method
Open Dental databases run on **MyISAM or InnoDB**. For a *file-level* backup to be consistent:
- A **hot copy** (copying the data directory while the service runs) can capture InnoDB mid-write → **corrupt, unrestorable** backup. Open Dental's built-in Backup tool and most "online" file backups **cannot even restore InnoDB**.
- A **cold copy** — stop the service so it flushes and closes cleanly, copy, restart — is **consistent for both engines**. This is the gold-standard local backup.
Two details that make or break a cold copy:
1. **Copy the *entire* data directory, not just `data\opendental\`.** InnoDB's shared tablespace and redo logs (`ibdata1`, `ib_logfile*`) live at the **root** of the data directory. Copy only the `opendental` subfolder and an InnoDB restore will fail.
2. **Verify the service actually stopped before copying.** If it won't stop, do **not** copy — you'd capture a live datadir.
The script below does both, and **always restarts the service** (even if the copy fails), so the practice is never left down.
---
## 1. Pre-flight
- Run **on the database server**, in an **elevated** PowerShell, **off-hours** — the copy causes downtime; Open Dental is unavailable on every workstation while the service is stopped.
- Make sure **no one is in Open Dental** (fully closed on all workstations).
- Have a `<DEST>` on a **different physical disk** (or UNC share) with enough free space for the data directory **+** images.
## 2. Run the cold backup (primary method)
From an **elevated** PowerShell on the server:
```
irm https://gitea.ivangodwin.com/ops/rb/raw/branch/main/scripts/od-db-backup.ps1 | iex
```
It will:
1. Auto-detect the `<DB_SERVICE>`, `<DATA_DIR>`, and `<IMAGES_DIR>` (prompting to confirm/override).
2. Report sizes and destination free space, then confirm before doing anything.
3. **Stop `<DB_SERVICE>` and verify it reached *Stopped*** — aborting the copy if it doesn't.
4. `robocopy` the **whole** data directory to `<DEST>\od-backup-<timestamp>\data`, then the images to `…\OpenDentImages`.
5. **Restart `<DB_SERVICE>`** in a `finally` block — this runs even if the copy fails or is interrupted.
6. Capture recovery aids (best-effort, outside the downtime window): a copy of **`FreeDentalConfig.xml`** and the exact **Open Dental + MySQL/MariaDB versions**.
7. Write `MANIFEST.txt` and `backup.log` into the backup folder.
### Manual equivalent (offline / if you can't fetch the script)
Elevated PowerShell, no one in Open Dental:
```
net stop <DB_SERVICE>
Get-Service <DB_SERVICE> # confirm Status = Stopped BEFORE copying
robocopy "<DATA_DIR>" "<DEST>\od-backup\data" /E /COPY:DAT /R:2 /W:5
robocopy "<IMAGES_DIR>" "<DEST>\od-backup\OpenDentImages" /E /COPY:DAT /R:2 /W:5
net start <DB_SERVICE>
Get-Service <DB_SERVICE> # confirm Status = Running
```
> Copy `<DATA_DIR>` itself (the parent of `opendental`), **not** `<DATA_DIR>\opendental` — you need `ibdata1` / `ib_logfile*` at the datadir root for InnoDB. If `net start` fails, start it immediately: the practice cannot work until the DB is back up.
## 3. Supplemental logical backup (mysqldump)
A cold copy is engine-perfect but physical — a portable **logical** dump is a valuable second line (survives binary corruption, restores to any server/version). It runs **while the service is up**, so schedule it separately (e.g. mid-day incremental in addition to the nightly cold copy). Slight slowness while it runs.
```
mysqldump -u <DB_USER> -p --single-transaction --quick --max-allowed-packet=1024M --default-character-set=utf8 --routines --events opendental > "<DEST>\opendental-<date>.sql"
```
- `--single-transaction` gives a consistent snapshot of **InnoDB** without locking the practice out. (For **MyISAM**, that flag does not guarantee consistency — use the cold copy as the source of truth.)
- Compress the `.sql` afterward; it shrinks dramatically.
- The dump does **not** include the images folder — always pair it with an `<IMAGES_DIR>` copy.
## 4. Schedule it (daily minimum)
Open Dental's floor is **at least one backup per day**; combine an automated nightly job with an off-site copy.
Register the cold backup as a nightly **Task Scheduler** job (runs off-hours, as SYSTEM/admin). Because `irm | iex` prompts interactively, schedule a **local copy** of the script with the paths baked in (keep that filled-in copy in the **private tier**, not here), e.g.:
```
schtasks /Create /TN "OD Nightly Cold Backup" /TR "powershell -NoProfile -ExecutionPolicy Bypass -File C:\ops\od-db-backup-local.ps1" /SC DAILY /ST 23:30 /RU SYSTEM /RL HIGHEST
```
## 5. Verify — a backup you haven't restored is a guess
- **Test-restore** to an **isolated** machine periodically (matching Open Dental + MySQL/MariaDB versions): `net stop`, rename the existing `opendental` folder aside, drop in the backup's `data\` contents, `net start`, launch Open Dental, spot-check patients/images. Full step-by-step with a health checklist: **[`od-backup-verify.md`](od-backup-verify.md)** (isolated Hyper-V test-restore).
- **Never restore over a live production database** — data loss is irreversible.
- Confirm the nightly job is actually producing dated folders and that they leave the building (off-site / immutable copy) — ransomware that reaches the server will reach on-line backups too.
## 6. Retention & off-site (3-2-1)
- **3** copies, **2** media, **1** off-site. The `<DEST>` disk alone is not a backup strategy.
- Keep several daily generations plus weekly/monthly rollups; prune old `od-backup-<timestamp>` folders on a schedule.
- These files contain **PHI** — encrypt at rest and in transit; restrict access. HIPAA applies.
### Point off-site tools at `<DEST>`, not at the live database
Any off-site/cloud replication (Duplicati → Backblaze B2, Veeam, `rclone`, Wasabi, etc.) must use **`<DEST>` as its source** — it backs up *this backup*. It must **not** re-run the stop/copy against the live `<DATA_DIR>`.
- The `od-backup-<timestamp>` folders under `<DEST>` are already a **consistent, cold copy** produced with the service cleanly stopped. Copying *them* off-site is a safe file copy — no service stop, no downtime, no consistency risk.
- Never let a naive file-sync tool crawl the live `C:\mysql\data` directly. A hot copy of a running InnoDB datadir is **corrupt and unrestorable** — the exact failure this runbook exists to avoid. Off-site tools have no idea they need to stop the service first; that's *this* script's job, done once, up front.
- **Sequence:** this cold backup runs first (nightly, off-hours) → the off-site job runs **after** it completes, sourcing `<DEST>`. Stagger the schedules (e.g. cold backup 23:30, off-site upload 01:00) so the upload never reads a half-written `od-backup-<timestamp>` folder. If your tool supports it, exclude any in-progress/partial folder or upload only completed timestamps.
- Let the off-site tool own its **own** encryption + retention on top — B2/Wasabi object lock or Duplicati's immutability/versioning gives you the ransomware-resistant, off-site copy of the 3-2-1 rule.
---
## Security note
Backup media and dumps hold the **entire practice's PHI**. Treat them as the crown jewels: encrypt the destination, lock down share permissions, and keep at least one copy **off-line/immutable** so ransomware can't encrypt your backups along with production. Don't store `<DB_PASSWORD>` in the scheduled command line — use a MySQL option file / limited account.
Note that the captured `FreeDentalConfig.xml` stores the MySQL password **obfuscated but reversible** on **direct-connection** setups. The backup already contains all PHI so this doesn't change the requirement — but it's one more reason the destination must be encrypted and access-controlled. (Middle Tier setups don't put DB credentials in that file — see `od-cfg-persist.md`.)
## References
- Open Dental manual — Backups (overview): <https://opendental.com/manual/backups.html>
- Open Dental manual — Manual Backups: <https://www.opendental.com/manual/backupsmanual.html>
- Open Dental manual — Backup Tool: <https://www.opendental.com/manual/backuptool.html>
- Open Dental manual — FreeDentalConfig.xml: <https://www.opendental.com/manual/freedentalconfig.html>
- Open Dental — InnoDB (backup implications): <https://www.opendental.com/site/mysqlinnodb.html>
- Open Dental manual — MySQL Data Directory Management: <https://opendental.com/manual/mysqlmanage.html>
- MySQL — `mysqldump`, `--single-transaction`: <https://dev.mysql.com/doc/refman/en/mysqldump.html>