Skip to content

Commit bd3652c

Browse files
aljoschaampagent
andcommitted
docs: Limit retention design update to a replica addendum
Amp-Thread-ID: https://ampcode.com/threads/T-01a0ceb4-cb71-751f-9239-4f04146413fd Co-authored-by: Amp <amp@ampcode.com>
1 parent a160220 commit bd3652c

2 files changed

Lines changed: 6 additions & 23 deletions

File tree

‎doc/developer/design/20260817_durable_object_hydration_history.md‎

Lines changed: 0 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -198,10 +198,6 @@ then, best effort is the honest description of a sampler.
198198

199199
## Retention
200200

201-
`hydration_history_retention_period` controls object history retention and defaults
202-
to 30 days. Replica history has a separate period, described in the
203-
[replica history design](20260827_durable_replica_hydration_history.md#retention-and-schema-evolution).
204-
205201
Retention is another OCC mutation: it subscribes to rows older than the cutoff and
206202
writes their retractions at the observed frontier. Collection applies the same
207203
cutoff, so a still-live log row cannot resurrect an episode retention just

‎doc/developer/design/20260827_durable_replica_hydration_history.md‎

Lines changed: 6 additions & 19 deletions
Original file line numberDiff line numberDiff line change
@@ -85,22 +85,9 @@ Rows currently have a populated `finished_at` and the status `hydrated`.
8585
Resource columns are nullable because the available kernel and filesystem
8686
observations depend on the replica platform.
8787

88-
## Retention and schema evolution
89-
90-
`replica_hydration_history_retention_period` defaults to 120 days (four months),
91-
independently of the object table's 30-day `hydration_history_retention_period`.
92-
Replica history grows with replicas and episodes, rather than objects times
93-
replicas times episodes. Its smaller size makes longer retention practical
94-
for capacity planning and hydration and resource-usage trends across releases.
95-
96-
Each table's collection and retention steps use its own cutoff. Replica history
97-
ages out by `finished_at`, in bounded batches through the shared retention path.
98-
A zero period expires all completed episodes and prevents their recollection.
99-
Disabling collection suspends retention for both tables.
100-
101-
The replica table commits to additive-only schema evolution that preserves
102-
existing rows. It is exempt from bootstrap reset and forced re-sharding, and
103-
capacity-planning history must not be discarded for routine schema
104-
changes. An exceptional shard replacement remains a deliberate escape hatch:
105-
remove the existing replacement assert and exemption together, and explicitly
106-
call out the loss of replica history in the release notes.
88+
## Addendum: separate replica retention
89+
90+
[SQL-714](https://linear.app/materializeinc/issue/SQL-714) introduced
91+
`replica_hydration_history_retention_period` after this design, with a default of
92+
120 days (four months). Object history retains its separate 30-day
93+
`hydration_history_retention_period`.

0 commit comments

Comments
 (0)