Hello, Milos Nikic, le lun. 08 sept. 2025 09:41:05 -0700, a ecrit: > I have actually experimenting with this is and for some reason we deadlock if > we don't stop at non consecutive blocks.
? What do you mean by "stop"? In your initial patch you would not completely stop, as in you would just call pending_blocks_write and continue piling blocks in pb. If you let pending_blocks_add do the pending_blocks_write call for you, there shouldn't be any difference? > I also did a bunch of testing like creating many directories and sparse files > hoping that I could get some big savings on the amount of times store_write > was > called but I got to say I didn't. > > So either my tests are not great, or there isn't much benefit to gain here by > batching these writes for the disk case. Did you make some statistics on the occurrence of consecutive blocks? I would be not that surprised that the numbers are not very high since the metadata are not very large. It'll still be useful to commit something working. Samuel
