postgres/timescaledb
1
0
Fork 0
mirror of https://github.com/timescale/timescaledb.git synced 2025-05-26 08:41:09 +08:00
Code Issues Packages Projects Releases Wiki Activity

Clarify the comment for invalidation threshold

Browse Source 
Explicitly state that the invalidation threshold is ignored by the transactions
that use an isolation level above READ COMMITTED.
This commit is contained in:
Aleksander Alekseev 2021-10-06 14:49:02 +03:00 committed by Aleksander Alekseev
parent 3729c50017
commit cb13754c91
1 changed files with 7 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
tsl/src/continuous_aggs/invalidation_threshold.c
View File

@ -33,7 +33,7 @@
* amplification in "hot" regions---typically the "head" of the table. The * amplification in "hot" regions---typically the "head" of the table. The
* presumption is that most inserts happen at recent time intervals, and those * presumption is that most inserts happen at recent time intervals, and those
* intervals will be invalid until writes move out of them. Therefore, it * intervals will be invalid until writes move out of them. Therefore, it
* isnt't worth writing invalidations in that region since it is presumed * isn't worth writing invalidations in that region since it is presumed
* out-of-date anyway. Further, although it is possible to refresh a * out-of-date anyway. Further, although it is possible to refresh a
* continuous aggregate in those "hot" regions, it will lead to partially * continuous aggregate in those "hot" regions, it will lead to partially
* filled buckets. Thus, refreshing those intervals is discouraged since the * filled buckets. Thus, refreshing those intervals is discouraged since the
@ -43,7 +43,7 @@
* the threshold but not after it. * the threshold but not after it.
* *
* The invalidation threshold is moved forward (and only forward) by refreshes * The invalidation threshold is moved forward (and only forward) by refreshes
* on continuous aggregates when it covers a window that streches beyond the * on continuous aggregates when it covers a window that stretches beyond the
* current threshold. The invalidation threshold needs to be moved in its own * current threshold. The invalidation threshold needs to be moved in its own
* transaction, with exclusive access, before the refresh starts to * transaction, with exclusive access, before the refresh starts to
* materialize data. This is to avoid losing any invalidations that occur * materialize data. This is to avoid losing any invalidations that occur
@ -57,6 +57,11 @@
* | * |
* invalidation threshold * invalidation threshold
* *
* Transactions that use an isolation level stronger than READ COMMITTED will
* not be able to "see" changes to the invalidation threshold that may have
* been made while they were running. Therefore, they always create records
* in the hypertable invalidation log. See the cache_inval_entry_write()
* implementation in tsl/src/continuous_aggs/insert.c
*/ */
typedef struct InvalidationThresholdData typedef struct InvalidationThresholdData