DEV Community

david
david

Posted on Originally published at woitzik.dev

NFS vs local-path: The SQLite Trap That Corrupted My S3 Metadata

#kubernetes #storage #homelab #debugging

Originally published at woitzik.dev

Disclosure: This post contains Amazon affiliate links (marked with *). If you buy through them, I earn a small commission at no extra cost to you. I only link gear I actually own and use daily.

Garage S3, my self-hosted S3-compatible object store, uses SQLite for its metadata database. One morning, Terraform state operations started failing with SQLITE_CORRUPT. The bucket metadata was gone. The terraform-state bucket, the Atlantis lock table, the Velero backup index — all of it stored in a SQLite file that was now corrupted.

The root cause: Garage was running on an NFS-backed PersistentVolume. NFS doesn't support the file-locking primitives that SQLite's WAL (Write-Ahead Logging) mode requires. Under concurrent access — Garage's metadata writer and a Velero backup reading the same database — the NFS lock delegation failed silently, and SQLite wrote to overlapping pages.

View the complete homelab infrastructure source on GitHub 🐙

The Storage Classes

My k3s cluster has two StorageClasses:

# NFS — for most workloads
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: nfs-client
provisioner: nfs-subdir-external-provisioner
parameters:
  server: 10.0.20.100
  path: /archive
reclaimPolicy: Retain

# Local-path — for embedded databases
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: local-path
provisioner: rancher.io/local-path
reclaimPolicy: Retain
Enter fullscreen mode Exit fullscreen mode

NFS (nfs-client) is the default. It's backed by a dedicated LXC running an NFS server on ZFS, providing storage that survives pod rescheduling — a pod on k3s-12 can access the same PVC as a pod on k3s-13 because the NFS server is independent of any specific node.

Local-path (local-path) pins the PV to whichever node created it. If the pod reschedules to a different node, the PVC is inaccessible until the pod returns to the original node. This is a limitation, but it's the right trade-off for certain workloads.

Why SQLite and NFS Don't Work

SQLite's WAL mode requires fcntl() file locks — specifically, F_SETLK (non-blocking lock) and F_SETLKW (blocking lock). These locks coordinate access between concurrent processes writing to the same database file.

NFS handles file locks differently:

  1. NFSv3: No native lock support. fcntl() calls return success but locks are local to the client — two NFS clients can both acquire an "exclusive" lock on the same file simultaneously.

  2. NFSv4: Has LOCK operations, but the lock delegation model introduces latency and failure modes that SQLite's tight locking loop doesn't tolerate. If the NFS server is slow to respond to a lock request, SQLite's default 5-second busy timeout can expire, causing the application to retry — and the retry can conflict with the lock held by another client.

  3. Kubernetes NFS provisioner: The nfs-subdir-external-provisioner uses NFSv4, but the lock delegation is handled by the NFS server's rpc.lockd daemon, which runs in a separate process space. Under concurrent load, lockd can lose track of which client holds which lock.

The result: SQLite thinks it has an exclusive lock, but another process (or the same process on a different connection) also has a lock. Both write to the database file. Pages overlap. The database corrupts.

The Garage Incident

Garage runs with two storage mounts:

# kubernetes/apps/garage/garage.yml
volumes:
  - name: data
    persistentVolumeClaim:
      claimName: garage-data      # NFS — bucket objects
  - name: meta
    persistentVolumeClaim:
      claimName: garage-meta      # local-path — SQLite metadata
Enter fullscreen mode Exit fullscreen mode

The data volume (bucket objects) is on NFS — fine, because S3 object storage doesn't use file locks for individual files. The meta volume (SQLite database) was also on NFS initially. This worked until a Velero backup and a Terraform state write happened simultaneously.

Velero reads Garage's S3 API to enumerate backup objects. Terraform reads the same database to verify state file existence. Both hit SQLite through Garage's metadata layer. Under NFS, the concurrent reads triggered the lock delegation failure, and SQLite corrupted pages 169–184 of db.sqlite.

The Recovery

SQLite has a .recover command that can extract data from a corrupted database:

# Dump recoverable data from corrupted SQLite
sqlite3 db.sqlite ".recover" > recovered.sql

# Recreate the database from the dump
sqlite3 db_clean.sqlite < recovered.sql
Enter fullscreen mode Exit fullscreen mode

The .recover command scanned every page of the corrupted file and extracted whatever data it could read. For pages 169–184 (the corrupted range), it found partial data — enough to reconstruct the bucket and key metadata, but not enough to guarantee referential integrity.

After recovery, the missing objects (terraform-state bucket and Atlantis lock key) had to be re-inserted manually via Python:

import sqlite3
import msgpack

conn = sqlite3.connect("db_clean.sqlite")
# Garage uses msgpack-encoded metadata with 'G2key'/'G2bkt' prefixes
# Re-insert the terraform-state bucket
conn.execute("INSERT INTO buckets (name, ...) VALUES (...)")
conn.commit()
Enter fullscreen mode Exit fullscreen mode

The data was restored, but the trust was gone. The database could corrupt again under the same conditions.

The Fix

Move every embedded database to local-path:

# Garage — meta volume moved to local-path
volumes:
  - name: meta
    persistentVolumeClaim:
      claimName: garage-meta      # NOW: local-path (was: nfs-client)
Enter fullscreen mode Exit fullscreen mode

The apps that need local-path:

App Database Why local-path
Garage S3 SQLite (metadata) File-locking requirements
Mealie SQLite (recipes) WAL mode + concurrent access
Home Assistant SQLite (state) Inotify-based DB writes
Authelia PostgreSQL (CNPG) CNPG manages its own PV

PostgreSQL (via CNPG) doesn't have the SQLite lock problem because it uses its own file locking, but CNPG requires local-path or a CSI driver that supports ReadWriteOnce — NFS's ReadWriteMany semantics can confuse CNPG's WAL archiving.

The Trade-off

local-path means the PVC is pinned to one node. If the pod reschedules, it loses access to the data. For Garage, this is acceptable: Garage runs on a single node, and if that node goes down, the S3 data is unavailable regardless (it's on the same host).

For databases that need HA (Postgres, Redis), the solution isn't local-path or NFS — it's a managed operator (CNPG for Postgres) that handles replication and failover independently of the storage layer.

The principle: if the application uses file-level locking (SQLite, BoltDB, LMDB), it goes on local-path. If it uses network-level locking (PostgreSQL, MySQL), it goes on NFS or a managed operator.

SQLite on NFS is the same failure mode as running SQLite on an SMB share in a Windows domain: the file-locking semantics are fundamentally incompatible. In Azure, this maps to Azure Files (SMB-backed) vs. Azure Disk (block storage). Azure Files supports SMB locks but has the same delegation latency issues under concurrent access — any application that needs tight file-level locking should use Azure Disks, not Azure Files. The principle is identical: embedded databases need local, low-latency storage with native file-locking support.

Designing Data-Intensive Applications* covers exactly this class of correctness assumption — what a storage layer actually guarantees about concurrent access versus what an application silently assumes it guarantees — in far more depth than a corrupted db.sqlite file teaches you in the moment.

Top comments (1)

Collapse
 
raknaos profile image
Raknaos •

This is the failure I never got tired of: NFS gives you a filesystem that looks shared and locks that only look global. Byte-range locks on many client configurations are honoured per client, so two pods on two nodes each believe they own the same SQLite lock, and SQLITE_CORRUPT is the loud outcome. The quiet one is WAL: a stale read of -wal and -shm gives you a database that looks internally consistent but is a blend of two generations, and it never errors, it just returns rows that were already overwritten.

Moving meta to local-path is the right trade — you buy correctness with reschedulability. Two things I would want on top if this were mine: does the meta pod pin to the node (so a reschedule blocks rather than starting a second writer on a fresh empty volume), and do you run an PRAGMA integrity_check after a Velero restore? A backup that captured the main file without the WAL loses the last commits silently, which is a worse surprise than the corruption that started this.