Database is currently being compacted error in database writer component

Direct answer

The "database is currently being compacted" error in the CouchDB database writer is a busy/contention signal, not evidence of database corruption. Compaction rewrites the database file in the background and can contend for disk I/O, page cache, and file descriptors, which raises fsync/commit latency and p99 write latency for some requests. There is no universal latency threshold that applies to every deployment. Treat the contention as problematic when sustained p95 or p99 write latency rises above your service SLO, or when write timeouts and request queue buildup coincide with compaction periods.

Likely explanation versus confirmed facts

Item	Status	Notes
Writer error means all writes are blocked for the whole compaction	Not confirmed	CouchDB generally keeps databases available by writing to a new file and swapping at the end; the error indicates contention, not a total write stop.
Compaction can increase write latency and queue depth	Established behavior	Impact varies by disk, write rate, scheduling, and CouchDB 3.x minor version.
Automatic compaction should be disabled permanently during writes	Not recommended	Indefinite disabling risks unbounded database and view file growth.

Steps for this case

1. Record baseline p50, p95, and p99 write latency and error rate while compaction is idle, under the same concurrent write load.
2. Repeat the measurement during compaction and compare the delta. A sustained increase beyond your SLO, or write timeouts that line up with compaction start and end, signals problematic contention.
3. Inspect CouchDB logs and node status around each error to confirm compaction start and end times. Verify the exact stat names for your installed CouchDB 3.x minor version, because names and defaults differ across releases.
4. Run iostat -x or an equivalent during idle and compaction periods. Compare device percent util, await, queue length, and file-descriptor usage. If these rise only during compaction and return to baseline afterward, the bottleneck is compaction-related rather than general disk saturation.
5. Pause or reschedule compaction on one node. If the writer error and latency spike stop, compaction contention is confirmed.

Compaction scheduling guidance

For sustained write workloads, do not disable compaction permanently. Schedule compaction in low-write windows, limit concurrent compactions, and use OS I/O priority where available. Re-enable automatic compaction afterward to prevent unbounded file growth. Adjust based on real-time I/O headroom only after correlating latency spikes with compaction windows in your own environment.

Missing diagnostic detail: What is your exact CouchDB 3.x minor version? Stat names, compaction defaults, and auto-compaction behavior differ across 3.x releases, so this determines which metrics to inspect before changing scheduling.

Verify these steps in staging before applying to production, because compaction can temporarily add I/O during active writes.