[scd] Change the subcription locking mechanism to a simple, shared lock.#1363
[scd] Change the subcription locking mechanism to a simple, shared lock.#1363the-glu wants to merge 1 commit intointeruss:masterfrom
Conversation
the-glu
force-pushed
the
simple_db_lock
branch
from
ede5c74 to
015fcaf
Compare
| ## The SCD global lock option | ||
|
|
||
| !!! danger | ||
| All DSS instances in a DSS pool must use the same value for this option. Mixing will result in dramatically lower performance. |
There was a problem hiding this comment.
IMO there isn't enough safeguards or mitigation that this does not happen. At least IMO we want some visibility, e.g. expose this configuration flag on the _aux interface?
cc @BenjaminPelletier
There was a problem hiding this comment.
From InterUSS weekly:
- solution to this should not take a large effort
- failing fast is OK
- addressing this through doc only is OK
- exposing through aux interface is OK
There was a problem hiding this comment.
I added the info in the aux interface
There was a problem hiding this comment.
Then, since it is exposed, maybe it would makes sense adding some test case in monitoring ensuring the values are consistent across the instances?
There was a problem hiding this comment.
Mmm I'm not sure it's worth the hassle of having a full validation for that, as we may have better solutions in the future (e.g. with crdb support dropped) and that will only affect performances, in 'high load' cases.
the-glu
force-pushed
the
simple_db_lock
branch
from
64207a4 to
bb6050b
Compare
mickmis
left a comment
mickmis
left a comment
There was a problem hiding this comment.
LGTM modulo downfrom entry
the-glu
force-pushed
the
simple_db_lock
branch
from
de42be0 to
a9043fc
Compare
|
NB: CI failure should be fixed with #1374 |
2 participants
This PR proposes a new method of locking subscriptions, by having a shared lock used for all subscriptions, instead of locking all rows.
See #1311 (comment) for graphs and details about the behavior change.
I added documentation in operations/performances, expecting #1362 to reuse it as well.