I still haven’t tested the scenario with a power loss, I’ll test that first and document my findings.
I guess the update process is the same like with the previous one?
Same update process, yes
tested both Vero and Rpi2 with the new build.
Have the mentioned INFO: sqlite-osmc: memory is not mapped yet and INFO: sqlite-osmc: memory mapped IO is enabled in both kodi.log
Vero 2m 41s
Rpi2 2m 3s
Looks like it has not made a difference then
Sam
I see what you are referring to now. To keep the transaction ACID, the writeback order is important. I don’t think there will be any reordering on NAND, especially with NOOP IO scheduler, but this is worth looking in to.
Sam
It made a huge different from the original code, but not from the first test build, just to be clear.
Rpi2:
Original code: 3m 48s
First test build: 2m 3s
Second test build: 2m 3s
Vero:
Original code: 5m 18s
First test build: 2m 45s
Second test build: 2m 41s
Yes, so it seems that memory mapped IO may not be configurable from within Kodi itself, which is unfortunate.
Sam
I’ve tested out the scenario with a power loss.
The database becomes corrupt and the folders that were scanned (at least partially) are not accessible any more.
Kodi file manager works fine, but videos->files-> folder doesn’t get you anywhere.
I’ll update to a new build and test it out, but I’m sure I’ll end up with the same results as @thansen_dk.
I’ve updated to the new version, the scanning time is the same, although I didn’t find INFO: sqlite-osmc: memory is not mapped yet and INFO: sqlite-osmc: memory mapped IO in kodi.log.
Tried the scenario for a power loss during scanning and the library was unaffected!
Did you enable debug logging?, if not you not get sqlite messages in your log.
If you did not see this, you’re not on the latest version or debug logging is disabled
I haven’t changed anything regarding power loss in the last build, but pulling the plug is always a risky venture.
Ok just enabled debug logging and found those lines!
There is definitely no speed gain and even with this version power loss causes database corruption.
I have also tested OE 5.95.5 regarding the power loss scenario.
OE doesn’t corrupt the database, like with OSMC I tested this several times.
So whatever speed improvements they applied are not dangerous for the database like the ones we have with OSMC now.
Yes, looks like we’ll need to revert these commits, so the performance will be back as it was before.
Sam
I’m not sure how’s OE achieving those speeds then.
They must have done something.
The speed improvements from the improved SQLite library will be included in the November update and these will not affect filesystem integrity as the previous tests did.
Sam
2 Likes
Awesome!