Do completely independent multipart uploads.
In distributed mode, a lock was held to merge each multipart
upload as it was added. This lock was highly contested and
retries are expensive (timewise) in distributed mode.
Instead, each part adds its metadata information uniquely.
This eliminates the per object lock required for each to merge.
The metadata is read back and merged by "CompleteMultipartUpload"
without locks when constructing final object.
Co-authored-by: Harshavardhana <harsha@minio.io>
In a streaming response, the client knows the size of a streamed
message but never checks the message size. Add the check to error
out if the response message is truncated.
Main motivation is move towards a common backend format
for all different types of modes in MinIO, allowing for
a simpler code and predictable behavior across all features.
This PR also brings features such as versioning, replication,
transitioning to single drive setups.
The main goal of this PR is to solve the situation where disks stop
responding to operations. This generally causes an FD build-up and
eventually will crash the server.
This adds detection of hung disks, where calls on disk get stuck.
We add functionality to `xlStorageDiskIDCheck` where it keeps
track of the number of concurrent requests on a given disk.
A total number of 100 operations are allowed. If this limit is reached
we will block (but not reject) new requests, but we will monitor the
state of the disk.
If no requests have been completed or updated within a 15-second
window, we mark the disk as offline. Requests that are blocked will be
unblocked and return an error as "faulty disk".
New requests will be rejected until the disk is marked OK again.
Once a disk has been marked faulty, a check will run every 5 seconds that
will attempt to write and read back a file. As long as this fails the disk will
remain faulty.
To prevent lots of long-running requests to mark the disk faulty we
implement a callback feature that allows updating the status as parts
of these operations are running.
We add a reader and writer wrapper that will update the status of each
successful read/write operation. This should allow fine enough granularity
that a slow, but still operational disk will not reach 15 seconds where
50 operations have not progressed.
Note that errors themselves are not enough to mark a disk faulty.
A nil (or io.EOF) error will mark a disk as "good".
* Make concurrent disk setting configurable via `_MINIO_DISK_MAX_CONCURRENT`.
* de-couple IsOnline() from disk health tracker
The purpose of IsOnline() is to ensure that we
reconnect the drive only when the "drive" was
- disconnected from network we need to validate
if the drive is "correct" and is the same drive
which belongs to this server.
- drive was replaced we have to format it - we
support hot swapping of the drives.
IsOnline() is not meant for taking the drive offline
when it is hung, it is not useful we can let the
drive be online instead "return" errors for relevant
calls.
* return errFaultyDisk for DiskInfo() call
Co-authored-by: Harshavardhana <harsha@minio.io>
Possible future Improvements:
* Unify the REST server and local xlStorageDiskIDCheck. This would also improve stats significantly.
* Allow reads/writes to be aborted by the context.
* Add usage stats, concurrent count, blocked operations, etc.
In a distributed setup, a DiskInfo REST call to an unformatted disk
returns an error with no disk information, such as the disk endpoint
URL, which is unexpected.
request.Form uses less memory allocation and avoids gorilla mux matching
with weird characters in parameters such as '\n'
- Remove Queries() to avoid matching
- Ensure r.ParseForm is called to populate fields
- Add a unit test for object names with '\n'
The httpStreamResponse should not return until CloseWithError has been called.
Instead keep track of write state and skip writing/flushing if an error has occurred.
Fixes#13743
Regression from #13597 (not released)
- Go might reset the internal http.ResponseWriter() to `nil`
after Write() failure if the go-routine has returned, do not
flush() such scenarios and avoid spurious flushes() as
returning handlers always flush.
- fix some racy tests with the console
- avoid ticker leaks in certain situations
Borrowed idea from Go's usage of this
optimization for ReadFrom() on client
side, we should re-use the 32k buffers
io.Copy() allocates for generic copy
from a reader to writer.
the performance increase for reads for
really tiny objects is at this range
after this change.
> * Fastest: +7.89% (+1.3 MiB/s) throughput, +7.89% (+1308.1) obj/s
deleting collection of versions belonging
to same object, we can avoid re-reading
the xl.meta from the disk instead purge
all the requested versions in-memory,
the tradeoff is to allocate a map to de-dup
the versions, allow disks to be read only
once per object.
additionally reduce the data transfer between
nodes by shortening msgp data values.
#11878 added "keepHTTPResponseAlive" to CreateFile requests.
The problem is that it will begin writing to the response before the
body is read after 10 seconds. This will abort the writes on the
client-side, since it assumes the server has received what it wants.
The proposed solution here is to monitor the completion of the body
before beginning to send keepalive pings.
Fixes observed high number of goroutines stuck in `io.Copy` in
`github.com/minio/minio/cmd.(*xlStorage).CreateFile` and
`(*storageRESTClient).CreateFile` stuck in `http.DrainBody`.
we will allow situations such as
```
a/b/1.txt
a/b
```
and
```
a/b
a/b/1.txt
```
we are going to document that this usecase is
not supported and we will never support it, if
any application does this users have to delete
the top level parent to make sure namespace is
accessible at lower level.
rest of the situations where the prefixes get
created across sets are supported as is.
Download files from *any* bucket/path as an encrypted zip file.
The key is included in the response but can be separated so zip
and the key doesn't have to be sent on the same channel.
Requires https://github.com/minio/pkg/pull/6
This is to ensure that there are no projects
that try to import `minio/minio/pkg` into
their own repo. Any such common packages should
go to `https://github.com/minio/pkg`
in setups with lots of drives the server
startup is slow, initialize all local drives
in parallel before registering with muxer.
this speeds up when there are multiple pools
and large collection of drives.
gracefully start the server, if there are other drives
available - print enough information for administrator
to notice the errors in console.
Bonus: for really large streams use larger buffer for
writes.
This is an optimization by reducing one extra system call,
and many network operations. This reduction should increase
the performance for small file workloads.
Current implementation heavily relies on readAllFileInfo
but with the advent of xl.meta inlined with data, we cannot
easily avoid reading data when we are only interested is
updating metadata, this leads to invariably write
amplification during metadata updates, repeatedly reading
data when we are only interested in updating metadata.
This PR ensures that we implement a metadata only update
API at storage layer, that handles updates to metadata alone
for any given version - given the version is valid and
present.
This helps reduce the chattiness for following calls..
- PutObjectTags
- DeleteObjectTags
- PutObjectLegalHold
- PutObjectRetention
- ReplicateObject (updates metadata on replication status)
For large objects taking more than '3 minutes' response
times in a single PUT operation can timeout prematurely
as 'ResponseHeader' timeout hits for 3 minutes. Avoid
this by keeping the connection active during CreateFile
phase.
For large objects taking more than '3 minutes' response
times in a single PUT operation can timeout prematurely
as 'ResponseHeader' timeout hits for 3 minutes. Avoid
this by keeping the connection active during CreateFile
phase.
There was an io.LimitReader was missing for the 'length'
parameter for ranged requests, that would cause client to
get truncated responses and errors.
fixes#11651
This PR refactors the way we use buffers for O_DIRECT and
to re-use those buffers for messagepack reader writer.
After some extensive benchmarking found that not all objects
have this benefit, and only objects smaller than 64KiB see
this benefit overall.
Benefits are seen from almost all objects from
1KiB - 32KiB
Beyond this no objects see benefit with bulk call approach
as the latency of bytes sent over the wire v/s streaming
content directly from disk negate each other with no
remarkable benefits.
All other optimizations include reuse of msgp.Reader,
msgp.Writer using sync.Pool's for all internode calls.
Use separate sync.Pool for writes/reads
Avoid passing buffers for io.CopyBuffer()
if the writer or reader implement io.WriteTo or io.ReadFrom
respectively then its useless for sync.Pool to allocate
buffers on its own since that will be completely ignored
by the io.CopyBuffer Go implementation.
Improve this wherever we see this to be optimal.
This allows us to be more efficient on memory usage.
```
385 // copyBuffer is the actual implementation of Copy and CopyBuffer.
386 // if buf is nil, one is allocated.
387 func copyBuffer(dst Writer, src Reader, buf []byte) (written int64, err error) {
388 // If the reader has a WriteTo method, use it to do the copy.
389 // Avoids an allocation and a copy.
390 if wt, ok := src.(WriterTo); ok {
391 return wt.WriteTo(dst)
392 }
393 // Similarly, if the writer has a ReadFrom method, use it to do the copy.
394 if rt, ok := dst.(ReaderFrom); ok {
395 return rt.ReadFrom(src)
396 }
```
From readahead package
```
// WriteTo writes data to w until there's no more data to write or when an error occurs.
// The return value n is the number of bytes written.
// Any error encountered during the write is also returned.
func (a *reader) WriteTo(w io.Writer) (n int64, err error) {
if a.err != nil {
return 0, a.err
}
n = 0
for {
err = a.fill()
if err != nil {
return n, err
}
n2, err := w.Write(a.cur.buffer())
a.cur.inc(n2)
n += int64(n2)
if err != nil {
return n, err
}
```
The only purpose of check-dir flag in
ReadVersion is to return 404 when
an object has xl.meta but without data.
This is causing an extract call to the disk
which can be penalizing in case of busy system
where disks receive many concurrent access.
Do listings with prefix filter when bloom filter is dirty.
This will forward the prefix filter to the lister which will make it
only scan the folders/objects with the specified prefix.
If we have a clean bloom filter we try to build a more generally
useful cache so in that case, we will list all objects/folders.
Similar to #10775 for fewer memory allocations, since we use
getOnlineDisks() extensively for listing we should optimize it
further.
Additionally, remove all unused walkers from the storage layer
WriteAll saw 127GB allocs in a 5 minute timeframe for 4MiB buffers
used by `io.CopyBuffer` even if they are pooled.
Since all writers appear to write byte buffers, just send those
instead and write directly. The files are opened through the `os`
package so they have no special properties anyway.
This removes the alloc and copy for each operation.
REST sends content length so a precise alloc can be made.
this reduces allocations in order of magnitude
Also, revert "erasure: delete dangling objects automatically (#10765)"
affects list caching should be investigated.
Design: https://gist.github.com/klauspost/025c09b48ed4a1293c917cecfabdf21c
Gist of improvements:
* Cross-server caching and listing will use the same data across servers and requests.
* Lists can be arbitrarily resumed at a constant speed.
* Metadata for all files scanned is stored for streaming retrieval.
* The existing bloom filters controlled by the crawler is used for validating caches.
* Concurrent requests for the same data (or parts of it) will not spawn additional walkers.
* Listing a subdirectory of an existing recursive cache will use the cache.
* All listing operations are fully streamable so the number of objects in a bucket no
longer dictates the amount of memory.
* Listings can be handled by any server within the cluster.
* Caches are cleaned up when out of date or superseded by a more recent one.
add a hint on the disk to allow for tracking fresh disk
being healed, to allow for restartable heals, and also
use this as a way to track and remove disks.
There are more pending changes where we should move
all the disk formatting logic to backend drives, this
PR doesn't deal with this refactor instead makes it
easier to track healing in the future.
It was observed in VMware vsphere environment during a
pod replacement, `mc admin info` might report incorrect
offline nodes for the replaced drive. This issue eventually
goes away but requires quite a lot of time for all servers
to be in sync.
This PR fixes this behavior properly.
Add context to all (non-trivial) calls to the storage layer.
Contexts are propagated through the REST client.
- `context.TODO()` is left in place for the places where it needs to be added to the caller.
- `endWalkCh` could probably be removed from the walkers, but no changes so far.
The "dangerous" part is that now a caller disconnecting *will* propagate down, so a
"delete" operation will now be interrupted. In some cases we might want to disconnect
this functionality so the operation completes if it has started, leaving the system in a cleaner state.
When crawling never use a disk we know is healing.
Most of the change involves keeping track of the original endpoint on xlStorage
and this also fixes DiskInfo.Endpoint never being populated.
Heal master will print `data-crawl: Disk "http://localhost:9001/data/mindev/data2/xl1" is
Healing, skipping` once on a cycle (no more often than every 5m).
this is to detect situations of corruption disk
format etc errors quickly and keep the disk online
in such scenarios for requests to fail appropriately.
- admin info node offline check is now quicker
- admin info now doesn't duplicate the code
across doing the same checks for disks
- rely on StorageInfo to return appropriate errors
instead of calling locally.
- diskID checks now return proper errors when
disk not found v/s format.json missing.
- add more disk states for more clarity on the
underlying disk errors.
- Add changes to ensure remote disks are not
incorrectly taken online if their order has
changed or are incorrect disks.
- Bring changes to peer to detect disconnection
with separate Health handler, to avoid a
rather expensive call GetLocakDiskIDs()
- Follow up on the same changes for Lockers
as well
- Implement a new xl.json 2.0.0 format to support,
this moves the entire marshaling logic to POSIX
layer, top layer always consumes a common FileInfo
construct which simplifies the metadata reads.
- Implement list object versions
- Migrate to siphash from crchash for new deployments
for object placements.
Fixes#2111
GetDiskID() in storage rest client does not really issue a REST request
to the remote disk, but returns an in-memory value instead.
However, GetDiskID() should return an error when format.json is not
found or for other similar issues (unmounted disks, etc..)
GetDiskID() is only called when formatting disks and getting storage
informatio, hence this commit should not have a performance degradation.
Shuffling arguments that we pass to MinIO server are supported. However,
when that happens, Prometheus returns wrong information about disks usage
and online/offline status.
The commit fixes the issue by avoiding relying on xl.endpoints since
it is not ordered.
The `keepHTTPResponseAlive` would cause errors to be
returned with status OK.
- Add '32' as a filler byte until a response is ready
- '0' to indicate the response is ready to be consumed
- '1' to indicate response has an error which needs
to be returned to the caller
Clear out 'file not found' errors from dir walker, since it may be
in a folder that has been deleted since it was scanned.
- Add conservative timeouts upto 3 minutes
for internode communication
- Add aggressive timeouts of 30 seconds
for gateway communication
Fixes#9105Fixes#8732Fixes#8881Fixes#8376Fixes#9028
Bulk delete API was using cleanupObjectsBulk() which calls posix
listing and delete API to remove objects internal files in the
backend (xl.json and parts) one by one.
Add DeletePrefixes in the storage API to remove the content
of a directory in a single call.
Also use a remove goroutine for each disk to accelerate removal.
Upgrades between releases are failing due to strict
rule to avoid rolling upgrades, it is enough to
bump up APIs between versions to allow for quorum
failure and wait times. Authentication failures are
catastrophic in nature which leads to server not
be able to upgrade properly.
Fixes#9021Fixes#8968
multi-delete API failed with write quorum errors
under following situations
- list of files requested for delete doesn't exist
anymore can lead to quorum errors and failure
- due to usage of query param for paths, for really
long paths MinIO server rejects these requests as
malformed as unexpected.
This was reproduced with warp
JWT parsing is simplified by using a custom claim
data structure such as MapClaims{}, also writes
a custom Unmarshaller for faster unmarshalling.
- Avoid as much reflections as possible
- Provide the right types for functions as much
as possible
- Avoid strings.Join, strings.Split to reduce
allocations, rely on indexes directly.
