- optimize writing part.N.meta by writing both part.N
and its meta in sequence without network component.
- remove part.N.meta, part.N which were partially success
ful, in quorum loss situations during renamePart()
- allow for strict read quorum check arbitrated via ETag
for the given part number, this makes it double safer
upon final commit.
- return an appropriate error when read quorum is missing,
instead of returning InvalidPart{}, which is non-retryable
error. This kind of situation can happen when many
nodes are going offline in rotation, an example of such
a restart() behavior is statefulset updates in k8s.
fixes#20091
deadlines per moveToTrash() allows for a more granular timeout
approach for syscalls, instead of an aggregate timeout.
This PR also enhances multipart state cleanup to be optimal by
removing 100's of multipart network rename() calls into single
network call.
instead upon any error in renameData(), we still
preserve the existing dataDir in some form for
recoverability in strange situations such as out
of disk space type errors.
Bonus: avoid running list and heal() instead allow
versions disparity to return the actual versions,
uuid to heal. Currently limit this to 100 versions
and lesser disparate objects.
an undo now reverts back the xl.meta from xl.meta.bkp
during overwrites on such flaky setups.
Bonus: Save N depth syscalls via skipping the parents
upon overwrites and versioned updates.
Flaky setup examples are stretch clusters with regular
packet drops etc, we need to add some defensive code
around to avoid dangling objects.
since mid 2018 we do not have any deployments
without deployment-id, it is time to put this
code to rest, this PR removes this old code as
its no longer valuable.
on setups with 1000's of drives these are all
quite expensive operations.
the disk location never changes in the lifetime of a
MinIO cluster, even if it did validate this close to the
disk instead at the higher layer.
Return appropriate errors indicating an invalid drive, so
that the drive is not recognized as part of a valid
drive.
Each Put, List, Multipart operations heavily rely on making
GetBucketInfo() call to verify if bucket exists or not on
a regular basis. This has a large performance cost when there
are tons of servers involved.
We did optimize this part by vectorizing the bucket calls,
however its not enough, beyond 100 nodes and this becomes
fairly visible in terms of performance.
- Move RenameFile to websockets
- Move ReadAll that is primarily is used
for reading 'format.json' to to websockets
- Optimize DiskInfo calls, and provide a way
to make a NoOp DiskInfo call.
NOTE: This feature is not retro-active; it will not cater to previous transactions
on existing setups.
To enable this feature, please set ` _MINIO_DRIVE_QUORUM=on` environment
variable as part of systemd service or k8s configmap.
Once this has been enabled, you need to also set `list_quorum`.
```
~ mc admin config set alias/ api list_quorum=auto`
```
A new debugging tool is available to check for any missing counters.
Bonus: allow replication to attempt Deletes/Puts when
the remote returns quorum errors of some kind, this is
to ensure that MinIO can rewrite the namespace with the
latest version that exists on the source.
.metacache objects are transient in nature, and are better left to
use page-cache effectively to avoid using more IOPs on the disks.
this allows for incoming calls to be not taxed heavily due to
multiple large batch listings.
xl.meta gets written and never rolled back, however
we definitely need to validate the state that is
persisted on the disk, if there are inconsistencies
- more than write quorum we should return an error
to the client
- if write quorum was achieved however there are
inconsistent xl.meta's we should simply trigger
an MRF on them
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>
avoids creating new transport for each `isServerResolvable`
request, instead re-use the available global transport and do
not try to forcibly close connections to avoid TIME_WAIT
build upon large clusters.
Never use httpClient.CloseIdleConnections() since that can have
a drastic effect on existing connections on the transport pool.
Remove it everywhere.
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.
- add checks such that swapped disks are detected
and ignored - never used for normal operations.
- implement `unrecognizedDisk` to be ignored with
all operations returning `errDiskNotFound`.
- also add checks such that we do not load unexpected
disks while connecting automatically.
- additionally humanize the values when printing the errors.
Bonus: fixes handling of non-quorum situations in
getLatestFileInfo(), that does not work when 2 drives
are down, currently this function would return errors
incorrectly.
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.
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
MRF does not detect when a node is disconnected and reconnected quickly
this change will ensure that MRF is alerted by comparing the last disk
reconnection timestamp with the last MRF check time.
Signed-off-by: Anis Elleuch <anis@min.io>
Co-authored-by: Klaus Post <klauspost@gmail.com>
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)
* Provide information on *actively* healing, buckets healed/queued, objects healed/failed.
* Add concurrent healing of multiple sets (typically on startup).
* Add bucket level resume, so restarts will only heal non-healed buckets.
* Print summary after healing a disk is done.
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.
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.
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.
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.
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.
Harshavardhana
Anis Eleuch
Aditya Manthramurthy
Klaus Post
Praveen raj Mani