-
Notifications
You must be signed in to change notification settings - Fork 8
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Restore and Check. #191
base: main
Are you sure you want to change the base?
Restore and Check. #191
Conversation
67bbd3c
to
ae4b55a
Compare
ae4b55a
to
a2364f2
Compare
a2364f2
to
fac2fec
Compare
pushed el7 fixes into this. |
Let's pull out the supporting commits that are fine on their own and don't really have much to do with parallel restore. We can land them in their own PR and then focus on fixing up the parallel restore commits. I think the following should do it: |
See: #201
|
Signed-off-by: Zach Brown <[email protected]> Signed-off-by: Auke Kok <[email protected]>
As I was committing the initial check command I had only partially completed a rename of the function that checks the metadata allocators. Signed-off-by: Zach Brown <[email protected]>
Signed-off-by: Zach Brown <[email protected]>
Generally as we call block_get() we should validate that if the block has a hdr, at a minimum the crc is correct and the magic value is the expected value passed, and the fsid matches the superblock. This function implements just that. Returns -EINVAL, up to the caller to report a problem() and handle the outcome. For now the code just hard fails, which incedentally makes it fail the clobber-repair.sh tests I wrote. Signed-off-by: Auke Kok <[email protected]>
Adds basic man page content for the `check` subcommand. Signed-off-by: Auke Kok <[email protected]>
We check superblock magic, crc, flags. data device superblock is checked but a little less thorough. We check whether the device is still mounted, since that would make checking invalid to begin with. Quorum blocks are validated to have sane contents. We add a global problem counter so we can trivially measure and report whether any problem was found at all, instead of iterating over all the problems and checking each individual count. We pick the standard exit code values from `fsck` and mirror their intentional behavior. This results in `fsck.scoutfs` can now be trivially created by making it a wrapper around `scoutfs check`. Signed-off-by: Auke Kok <[email protected]> Signed-off-by: Hunter Shaffer <[email protected]>
Signed-off-by: Zach Brown <[email protected]> Signed-off-by: Hunter Shaffer <[email protected]> Signed-off-by: Auke Kok <[email protected]>
This is the benchmark binary that bulk creates filesystem items, xattrs and is heavily threaded to scope the performance of the library. The test script invokes it to validate some basic constraints. Signed-off-by: Zach Brown <[email protected]> Signed-off-by: Hunter Shaffer <[email protected]> Signed-off-by: Auke Kok <[email protected]>
This tool compies a source tree (whether it's scoutfs or not) into an offline scoutfs meta device. It has only those 2 parameters and does a single-process walk of the tree to restore all items while preservice as much of the metadata as possible. Signed-off-by: Hunter Shaffer <[email protected]> Signed-off-by: Auke Kok <[email protected]>
92c433c
to
edacd17
Compare
rebased onto |
This is failing tests on |
Yeah, the test shouldn't be letting the raw df output through to be compared. |
This branch against
main
contains both the parallel restore library, tests, as well as the check code.The check code has been reduced to our validation tool to identify inconsistencies in an umounted scoutfs meta device, only. All error injection and repair code has been omitted. The check code is used to validate the output of the parallel restore test cases.
scoutfs check
or there is an error inserting duringrestore
.New issues from @chaowang-versity:
1
where that number should be larger than1
for obvious reasons.util.h
should be exported, becauseformat.h
usesDIV_ROUND_UP
which is defined in it.