This PR implements an object layer which
combines input erasure sets of XL layers
into a unified namespace.
This object layer extends the existing
erasure coded implementation, it is assumed
in this design that providing > 16 disks is
a static configuration as well i.e if you started
the setup with 32 disks with 4 sets 8 disks per
pack then you would need to provide 4 sets always.
Some design details and restrictions:
- Objects are distributed using consistent ordering
to a unique erasure coded layer.
- Each pack has its own dsync so locks are synchronized
properly at pack (erasure layer).
- Each pack still has a maximum of 16 disks
requirement, you can start with multiple
such sets statically.
- Static sets set of disks and cannot be
changed, there is no elastic expansion allowed.
- Static sets set of disks and cannot be
changed, there is no elastic removal allowed.
- ListObjects() across sets can be noticeably
slower since List happens on all servers,
and is merged at this sets layer.
Fixes#5465Fixes#5464Fixes#5461Fixes#5460Fixes#5459Fixes#5458Fixes#5460Fixes#5488Fixes#5489Fixes#5497Fixes#5496
- Changes related to moving admin APIs
- admin APIs now have an endpoint under /minio/admin
- admin APIs are now versioned - a new API to server the version is
added at "GET /minio/admin/version" and all API operations have the
path prefix /minio/admin/v1/<operation>
- new service stop API added
- credentials change API is moved to /minio/admin/v1/config/credential
- credentials change API and configuration get/set API now require TLS
so that credentials are protected
- all API requests now receive JSON
- heal APIs are disabled as they will be changed substantially
- Heal API changes
Heal API is now provided at a single endpoint with the ability for a
client to start a heal sequence on all the data in the server, a
single bucket, or under a prefix within a bucket.
When a heal sequence is started, the server returns a unique token
that needs to be used for subsequent 'status' requests to fetch heal
results.
On each status request from the client, the server returns heal result
records that it has accumulated since the previous status request. The
server accumulates upto 1000 records and pauses healing further
objects until the client requests for status. If the client does not
request any further records for a long time, the server aborts the
heal sequence automatically.
A heal result record is returned for each entity healed on the server,
such as system metadata, object metadata, buckets and objects, and has
information about the before and after states on each disk.
A client may request to force restart a heal sequence - this causes
the running heal sequence to be aborted at the next safe spot and
starts a new heal sequence.
Under any concurrent removeObjects in progress
might have removed the parents of the same prefix
for which there is an ongoing putObject request.
An inconsistent situation may arise as explained
below even under sufficient locking.
PutObject is almost successful at the last stage when
a temporary file is renamed to its actual namespace
at `a/b/c/object1`. Concurrently a RemoveObject is
also in progress at the same prefix for an `a/b/c/object2`.
To create the object1 at location `a/b/c` PutObject has
to create all the parents recursively.
```
a/b/c - os.MkdirAll loops through has now created
'a/' and 'b/' about to create 'c/'
a/b/c/object2 - at this point 'c/' and 'object2'
are deleted about to delete b/
```
Now for os.MkdirAll loop the expected situation is
that top level parent 'a/b/' exists which it created
, such that it can create 'c/' - since removeObject
and putObject do not compete for lock due to holding
locks at different resources. removeObject proceeds
to delete parent 'b/' since 'c/' is not yet present,
once deleted 'os.MkdirAll' would receive an error as
syscall.ENOENT which would fail the putObject request.
This PR tries to address this issue by implementing
a safer/guarded approach where we would retry an operation
such as `os.MkdirAll` and `os.Rename` if both operations
observe syscall.ENOENT.
Fixes#5254
Apache Spark sends getObject requests with trailing "/".
This PR updates the getObjectInfo to stat for files
even if they are sent with trailing "/".
Fixes#2965
Verify() was being called by caller after the data
has been successfully read after io.EOF. This disconnection
opens a race under concurrent access to such an object.
Verification is not necessary outside of Read() call,
we can simply just do checksum verification right inside
Read() call at io.EOF.
This approach simplifies the usage.
Every so often we get requirements for creating
directories/prefixes and we end up rejecting
such requirements. This PR implements this and
allows empty directories without any new file
addition to backend.
Existing lower APIs themselves are leveraged to provide
this behavior. Only FS backend supports this for
the time being as desired.
Amazon S3 API expects all incoming stream has a content-length
set it was superflous for us to support object layer which supports
unknown sized stream as well, this PR removes such requirements
and explicitly error out if input stream is less than zero.
It can happen that an incoming PutObject() request might
have inputs of following form eg:-
- bucketName is 'testbucket'
- objectName is '/'
bucketName exists and was previously created but there
are no other objects in this bucket. In a situation like
this parentDirIsObject() goes into an infinite loop.
Verifying that if '/' is an object fails on both backends
but the resulting `path.Dir('/')` returns `'/'` this causes
the closure to loop onto itself.
Fixes#4940
This change refactor the ObjectLayer PutObject and PutObjectPart
functions. Instead of passing an io.Reader and a size to PUT operations
ObejectLayer expects an HashReader.
A HashReader verifies the MD5 sum (and SHA256 sum if required) of the object.
This change updates all all PutObject(Part) calls and removes unnecessary code
in all ObjectLayer implementations.
Fixes#4923
Current code was just using io.ReadAll() on an fd()
which might have moved underneath due to a concurrent
read operation. Subsequent read will result in EOF
We should always seek back and read again. pread()
is allowed on all platforms use io.SectionReader to
read from the beginning of the file.
Fixes#4842
Since go1.8 os.RemoveAll and os.MkdirAll both support long
path names i.e UNC path on windows. The code we are carrying
was directly borrowed from `pkg/os` package and doesn't need
to be in our repo anymore. As a side affect this also
addresses our codecoverage issue.
Refer #4658
Under the call flow
```
Readdir
+
|
|
| path-entry
|
|
v
StatDir
```
Existing code was written in a manner where say
a bucket/top-level directory was indeed deleted
between Readdir() and before StatDir() we would
ignore certain errors. This is not a plausible
situation and might not happen in almost all
practical cases. We do not have to look for
or interpret these errors returned by StatDir()
instead we can just collect the successful
values and return back to the client. We do not
need to pre-maturely decide on bucket access
we just let filesystem decide subsequently for
real I/O operations.
Refer #4658
Looks like if we follow pattern such as
```
_ = rlk
```
Go can potentially kick in GC and close the fd when
the reference is lost, only speculation is that
the cause here is `SetFinalizer` which is set on
`os.close()` internally in `os` stdlib.
This is unexpected and unsual endeavour for Go, but
we have to make sure the reference is never lost
and always dies with the server.
Fixes#4530
This patch also reverts previous changes which were
merged for migration to the newer disk format. We will
be bringing these changes in subsequent releases. But
we wish to add protection in this release such that
future release migrations are protected.
Revert "fs: Migration should handle bucketConfigs as regular objects. (#4482)"
This reverts commit 976870a391.
Revert "fs: Migrate object metadata to objects directory. (#4195)"
This reverts commit 76f4f20609.
Current code failed to anticipate the existence of files
which could have been created to corrupt the namespace such
as `policy.json` file created at the bucket top level.
In the current release creating such as file conflicts
with the namespace for future bucket policy operations.
We implemented migration of backend format to avoid situations
such as these.
This PR handles this situation, makes sure that the
erroneous files should have been moved properly.
Fixes#4478
Currently even when bucket doesn't exist we wrongly
return success, when an object is a directory prefix with
'/' as suffix and is of size 0.
This PR fixes this behavior.
This PR also does backend format change to 1.0.1
from 1.0.0. Backward compatible changes are still
kept to read the 'md5Sum' key. But all new objects
will be stored with the same details under 'etag'.
Fixes#4312
This change adopts the upstream fix in this regard at
https://go-review.googlesource.com/#/c/41834/ for Minio's
purposes.
Go's current os.Stat() lacks support for lot of strange
windows files such as
- share symlinks on SMB2
- symlinks on docker nanoserver
- de-duplicated files on NTFS de-duplicated volume.
This PR attempts to incorporate the change mentioned here
https://blogs.msdn.microsoft.com/oldnewthing/20100212-00/?p=14963/
The article suggests to use Windows I/O manager to
dereference the symbolic link.
Fixes#4122
This is to comply with S3 behavior, we previously removed
reading `fs.json` for optimization reasons but we have a
reason to believe that providing ETag and using gjson
provides needed benefit of not having to deal with
unmarshalling overhead of golang stdlib.
Fixes#4028
`disksUnavailable` healStatus constant indicates that a given object
needs healing but one or more of disks requiring heal are offline. This
can be used by admin heal API consumers to distinguish between a
successful heal and a no-op since the outdated disks were offline.
HEAD Object for FS and XL was returning invalid object name when
an object name has a trailing slash separator, this PR changes the
behavior and will always return 404 object not found, this guarantees
a better compatibility with S3 spec.
* fs: Rename tempObjPath variable in fsCreateFile()
* fs/posix: Factor checkDiskFree() function
* fs: Add disk free check in fsCreateFile()
* posix: Move free disk check to createFile()
* xl: Relax free disk check in POSIX initialization
* fs: checkDiskFree checks for space to store data
This PR is for readability cleanup
- getOrderedDisks as shuffleDisks
- getOrderedPartsMetadata as shufflePartsMetadata
Distribution is now a second argument instead being the
primary input argument for brevity.
Also change the usage of type casted int64(0), instead
rely on direct type reference as `var variable int64` everywhere.
Make sure to skip reserved bucket names in `ListBuckets()`
current code didn't skip this properly and also generalize
this behavior for both XL and FS.
Harshavardhana