Notesreplication🌱 seedling
Read-Your-Writes Consistency
Also: Read-After-Write Consistency · Read Your Own Writes · Read My Writes
A session guarantee: after you write something, your later reads always reflect it. It promises nothing about other users' writes.
Why it exists
With asynchronous followers, you post a comment to the leader, reload, hit a lagging follower, and your comment is gone. Users read that as data loss.
How it works
- Route by ownership: read things the user can modify (their own profile) from the leader; everything else from followers.
- Route by recency: read from the leader for a window (say, a minute) after the user's last write, or skip followers lagging more than that.
- Position tokens: the client remembers its last write's log position (a PostgreSQL LSN, a MySQL GTID); a follower serves the read only once it has replayed past it.
- Wait at commit: PostgreSQL
synchronous_commit = remote_applyreturns only after synchronous standbys applied the write.
Trade-offs
| Technique | Cost |
|---|---|
| read from leader | leader load; defeats read scaling if most things are editable |
| position tokens | client or session must carry state; reads may wait |
| remote_apply | higher commit latency |
In practice
- Cross-device read-your-writes (write on phone, read on laptop) needs the token on the server, and both devices routed to the same region.
Pitfalls
- Remembering a wall-clock timestamp instead of a log position: clock skew makes it unreliable.
Connections
- Part of
- Replication Lag
Recall
These cards are also in your review deck, scheduled with spaced repetition.