Before this change writing with the all policy deadlocked while
uploading.
This change fixes the problem by fixing the multi reader, closing the
pipes at the correct time with the correct error. This is factored
into a new function as it was used twice.
This patch also adds a new test which tests the all policies.
Running all the tests for 1fichier takes too long due to the directory
reading rate limiter.
The backend tests do complete in a reasonable time (21 mins).
Before this fix we were reading the hash from the upload using the
string "ETag", however the go runtime normalises the tag into "Etag"
so we were in fact always reading an empty string.
This bug was introduced in
aeea4430d5 swift: efficiency: slim Object and reduce requests on upload
It was spotted by the integration tests.
The fix was just to use the canonical form "Etag" instead of "ETag".
In this commit
a2afa9aadd fs: Add directory to optional Purge interface
We failed to encrypt the directory name so the Purge failed.
This was spotted by the integration tests.
From now on the betas will be numbered for the version that they will
become, so:
v1.53.0-beta.NNNN.CCCCC
Where N is commit number and C is commit. When released this will
become v1.53.0 and the beta will become v1.54.0-beta.NNN.CCCCC.
The commit number is the count of the commits since the root of the
tree since we can no longer use the the git version numbers since the
last tag.
This will simplify building the stable branch but that release
procedure hasn't been revised yet.
This commit also injects the name of the branch for the beta builds
into the download path.
In this commit:
cbf3d43561 drive: fix missing items when listing using --fast-list / ListR
We introduced a bug where under specific circumstances it could cause
a "panic: send on closed channel".
This was caused by:
- rclone engaging the workaround from the commit above
- one of the listing routines returning an error
- this caused the `in` channel to be closed to stop the readers
- however the workaround was recycling stuff into the `in` channel at the time
- hence the panic on closed channel
This fix factors out the sending to the `in` channel into `sendJob`
and calls this both from the master go routine and the list
runners. `sendJob` detects the `in` channel being closed properly and
also deals correctly with contention on the `in` channel.
Fixes#4511
This patch provides the support of synchronous cache space recovery
to allow read threads to recover from ENOSPC errors when cache space
can be recovered from cache items that are not in use or safe to be
reset/emptied .
The patch complements the existing cache cleaning process in two ways.
Firstly, the existing cache cleaning process is time-driven that runs
periodically. The cache space can run out while the cache cleaner
thread is still waiting for its next scheduled run. The io threads
encountering ENOSPC return an internal error to the applications
in this case even when cache space can be recovered to avoid this
error. This patch addresses this problem by having the read threads
kick the cache cleaner thread in this condition to recover cache
space preventing unnecessary ENOSPC errors from being seen by the
applications.
Secondly, this patch enhances the cache cleaner to support cache
item reset. Currently the cache purge process removes cache
items that are not in use. This may not be sufficient when the
total size of the working set exceeds the cache directory's
capacity. Like in the current code, this patch starts the purge
process by removing cache files that are not in use. Cache items
whose access times are older than vfs-cache-max-age are removed first.
After that, other not-in-use items are removed in LRU order until
vfs-cache-max-size is reached. If the vfs-cache-max-size (the quota)
is still not reached at this time, this patch adds a cache reset
step to reset/empty cache files that are still in use but not
dirtied. This enables application processes to continue without
seeing an error even when the working set depletes the cache space
as long as there is not a large write working set hoarding the
entire cache space.
By design this patch does not add ENOSPC error recovery for write
IOs. Rclone does not empty a write cache item until the file data
is written back to the backend upon close. Allowing more cache
space to be consumed by dirty cache items when the cache space is
already running low would increase the risk of exhausting the cache
space in a way that the vfs mount becomes unreadable.
When using `rclone authorize` the hostname doesn't get set in the
config file.
This commit allows it to be set in the configurator and gives the user
a hint that it needs setting.
This reverts part of
151f03378f s3: fix upload of single files into buckets without create permission
This erroneously assumed that a HEAD request on a non existent object
would return "NotFound" if the bucket was found. In fact it returns
"NotFound" when the bucket isn't found also.
This will break the fix for #4297 - however that can be made to work
using the new --s3-assume-bucket-exists flag
Before this change, rclone was looking for the file without the
extension to see if it existed which meant that it never did.
This change checks the destination file exists firsts, before removing
the extension.
Google drive appears to no longer be copying the modification time of
google docs.
Setting the mod time immediately after the copy doesn't work either,
so this patch copies the object, waits for 1 second and then sets the
modtime.
Fixes#4517
This was only working for files in the root directory and wasn't
looking at the encoding.
This is fixed to use NewObject which takes both things into account
and it makes the share by ID instead of by path.
This problem was spotted by the integration tests.
Before this change we set the modtime of the cache file when all
writers had finished.
This has the unfortunate effect that the file is uploaded with the
wrong modtime which means on backends which can't set modtimes except
when uploading files it is wrong.
This change sets the modtime of the cache file immediately in the
cache and in turn sets the modtime in the file info.
This is what I wrote to Digital Ocean support on July 10, 2020 - alas
it didn't result in the rate limits dropping, so reluctantly I'm going
to remove DO from the integration tests since they never pass and have
no hope of ever passing while this rate limit is in effect.
----
Somewhere towards the end of June 2020 or the start of July 2020 my
integration tests between rclone ( https://rclone.org ) and Digital
Ocean started failing.
I tried moving the tests to different regions (currently they are
using AMS1 because I'm in Europe) with no improvement.
Rclone seems to be hitting this rate limit as documented here:
https://www.digitalocean.com/docs/spaces/#limits
- 2 COPYs per 5 minutes on any individual object in a Space
Rclone creates small objects about 100 bytes in size and renames them
a few times - this involves using the COPY call as S3 does not have a
rename API. The tests do this more than twice per object so hit the 5
minute timeout I think. Rclone does exponential backoff and fails
after 10 retries not having reached 5 minutes delay after 10 retries.
Having a 5 minute lockout on an S3 compatible API is surprising!
Rclone integration tests with about 30 other providers, none of which
have a rate limit like this.
I understand the need for a COPY rate limit as server side copying
large files can be resource intensive. However a 5 minute lockout for
copying 100 byte files seems excessive!
Might I humbly suggest that you reduce or eliminate this rate limit
for small files?
----
This was the reply
Unfortunately it is not possible to raise this limit or remove it
currently on our platform. I do see how this would interfere with type
of applications that need to copy many small files and will be happy
to take the feedback to our engineering team to see how we can improve
the spaces system in the future
Before this change we errored out if one upstream errored in Purge or
About.
This change checks for fs.ErrorDirNotFound and skips that backend in
this case.
