A warehouse’s durability is ClickHouse’s own, and you choose it per session or per query.
Inserts into a replicated table
By default (insert_quorum = 0) an insert is acknowledged once one replica of the shard has written
it; the shard’s other replicas then fetch it.
To wait until more replicas have it, set insert_quorum:
INSERT INTO events.page_views SETTINGS insert_quorum = 2 VALUES (now(), 42, '/');| Replicas | insert_quorum = 2 |
|---|---|
| 2 | Each insert waits for both replicas, and fails while one of them is down. |
| 3 | Each insert waits for two of the three, and keeps working while one is down. |
insert_quorum_timeout sets how long an insert waits.
Inserts through a Distributed table
By default an insert into a Distributed table is acknowledged once the node you’re connected to has
queued it; that node then sends each row to its shard. With insert_distributed_sync = 1 (also named
distributed_foreground_insert), the insert
waits until every shard has written its rows, and with insert_quorum set, until the quorum has.
INSERT INTO events.page_views_all SETTINGS insert_distributed_sync = 1, insert_quorum = 2
VALUES (now(), 42, '/');For a session or a query
A user can SET these for its session, or a client can send them with each query (on HTTPS, as URL
parameters such as ?insert_quorum=2).