search

ecdye / zram-config

A complete zram-config utility for swap, directories, and logs to reduce SD, NAND and eMMC block wear.
MIT License
447 stars 54 forks source link

zram dir read-only #13

Closed mstormi closed 5 years ago

mstormi commented 5 years ago

I have one zram directory that cannot be written to (read only errors) dir lz4 50M 300M /var/lib/openhab2 /openhab2.bind The others work fine meanwhile. log zstd 50M 250M /var/log /log.bind /volatile/oldlogs dir lz4 10M 1000M /var/cache /cache.bind

Any idea ? I think it was writable for some time in the beginning after boot. Eventually the difference is that the amount of data written is larger than the mem_limit - could that be an explanation ?

StuartIanNaylor commented 5 years ago

mem_limit is uncompressed actually memory use. As long as the it managed to compress into the mem_limit approx x3 then it should fit.

I think its overlayfs and I have stopped being active for a couple of months and much is forgotten. I got the feeling in 4.17 or 4.18 a load of overlayfs patches went into the kernel and thought 4.19 was imminent and never bothered, but 4.19 and testing has taken much longer than I expected. I think its that or I am doing something stupid with the initial bind.

mstormi commented 5 years ago

mem_limit of the incriminated directory x3 = 150M was still less than that directory took on disk. Most of it is recreated on openHAB start so I believe the amount of data written could have surpassed that limit. I encountered this on a second occasion (the /var/cache shown above) whose mem_limit was obviously way too low, and it started showing after apt-get update (the probably only program to result in writes to that dir). I've now increased mem_limit of the first dir and removed the /var/cache line. No problems since. FWIW, I'm using kernel 4.14.98. Will keep you updated here.

Will you keep supporting/keep this code up to date ? I've just created this PR to use your wonderful creation in openHABian. Please keep up the great work.

Another thing, could you please explain how the syncing and move to oldlogs mount works ? Most important thing I'm asking myself is how often the sync will take place and when rotated logs will be moved to oldlogs. A pointer to existing docs will suffice.

StuartIanNaylor commented 5 years ago

Its a compromise of gaining the CoW copy up on write of OverlaysFS which always confuses me as it does have kernel inclusion but in comparison to Aufs has a near lack of tools. It uses https://github.com/kmxz/overlayfs-tools merge down but requires the overlayfs to be remounted as its likely the merge layer upper & lower will become out of sync. You gain huge reductions in memory usage as only writes are pulled in zram but syncing needs to be offline. I have asked the overlayfs guys a hey "you supplying any tools guys?" but the answer seems to be no. I have thought about maybe doing a Aufs version.