Syncthing: Reconsider db check on upgrade
It used to be quite quick and painless; seconds on my computer. Now up to about five minutes on my computer. Other setups are worse:
<sphalerite> Is there any way to make "Checking db due to upgrade" faster? It takes hours on my phone.
I think it's the block loading that now happens...
calmh
>All comments
I do believe it's been painfully slow even before loading blocks - checking all the global entries is terribly slow in itself. Until now there's not a single "device present in global..." panic in v1.9.0, so I am hopeful https://github.com/syncthing/syncthing/pull/6888 fixed that for good. Unconditionally repairing sequences feels unnecessary anyway, as when sequence problems do occur, we detect and fix those (only if necessary) both when loading the db and when sending indexes. Given weird metadata issues popping up all the time I'd keep the meta recalculation on upgrade for now. If sentry stays clean on the global issue, I'd definitely drop it for the next release.
imsodin
on 24 Aug 2020
Related issues
da2x
路
3Comments
kakra
路
3Comments
tomasz1986
路
3Comments
trymeouteh
路
4Comments
gabriel-fallen
路
4Comments
Most helpful comment
I do believe it's been painfully slow even before loading blocks - checking all the global entries is terribly slow in itself. Until now there's not a single "device present in global..." panic in v1.9.0, so I am hopeful https://github.com/syncthing/syncthing/pull/6888 fixed that for good. Unconditionally repairing sequences feels unnecessary anyway, as when sequence problems do occur, we detect and fix those (only if necessary) both when loading the db and when sending indexes. Given weird metadata issues popping up all the time I'd keep the meta recalculation on upgrade for now. If sentry stays clean on the global issue, I'd definitely drop it for the next release.