Persistence

RDB snapshots — save, load, and crash recovery.

How It Works

FyroDB uses RDB snapshots — the same persistence model as Redis.

  • On startup — loads fyrodb.rdb from the current directory (if it exists)
  • Every 5 minutes — automatic background save (configurable via FYRODB_RDB_INTERVAL)
  • On shutdown — saves on SIGTERM or Ctrl+C after draining client writes
  • Manual — BGSAVE command triggers an immediate background save

Atomic Writes

Saves are crash-safe:

  1. Write to temporary file (fyrodb.rdb.tmp)
  2. fsync the file (flush to disk)
  3. rename temp → final (atomic on POSIX)
  4. fsync the directory

A crash mid-save never corrupts the existing snapshot.

Non-Blocking Save

The save iterates slot-by-slot with brief EBR pins. Between each slot, the guard is released — garbage collection proceeds normally. No multi-second GC stalls during saves.

File Format

[FLDB magic 4 bytes] [version 1 byte]
[type 1b] [ttl_ms 8b] [key_len 4b] [key] [value_payload]
[type 1b] [ttl_ms 8b] [key_len 4b] [key] [value_payload]
...
[0xFF EOF marker]
  • Strings: type=0 + length-prefixed string value
  • Hashes: type=1 + field count + (field + value) pairs

Load Behavior

  • Expired keys are skipped without allocation
  • String lengths capped at 512MB (rejects corrupted files)
  • Hash field counts capped at 10M
  • If EOF marker is missing, loads available data with a warning

Configuration

FYRODB_RDB_PATH=fyrodb.rdb       # snapshot file location
FYRODB_RDB_INTERVAL=300          # auto-save interval (seconds)

Disable Persistence

Set a very large interval to effectively disable auto-save:

FYRODB_RDB_INTERVAL=999999999

Manual saves via BGSAVE still work.