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.
This PR adds a WebSocket grid feature that allows servers to communicate via
a single two-way connection.
There are two request types:
* Single requests, which are `[]byte => ([]byte, error)`. This is for efficient small
roundtrips with small payloads.
* Streaming requests which are `[]byte, chan []byte => chan []byte (and error)`,
which allows for different combinations of full two-way streams with an initial payload.
Only a single stream is created between two machines - and there is, as such, no
server/client relation since both sides can initiate and handle requests. Which server
initiates the request is decided deterministically on the server names.
Requests are made through a mux client and server, which handles message
passing, congestion, cancelation, timeouts, etc.
If a connection is lost, all requests are canceled, and the calling server will try
to reconnect. Registered handlers can operate directly on byte
slices or use a higher-level generics abstraction.
There is no versioning of handlers/clients, and incompatible changes should
be handled by adding new handlers.
The request path can be changed to a new one for any protocol changes.
First, all servers create a "Manager." The manager must know its address
as well as all remote addresses. This will manage all connections.
To get a connection to any remote, ask the manager to provide it given
the remote address using.
```
func (m *Manager) Connection(host string) *Connection
```
All serverside handlers must also be registered on the manager. This will
make sure that all incoming requests are served. The number of in-flight
requests and responses must also be given for streaming requests.
The "Connection" returned manages the mux-clients. Requests issued
to the connection will be sent to the remote.
* `func (c *Connection) Request(ctx context.Context, h HandlerID, req []byte) ([]byte, error)`
performs a single request and returns the result. Any deadline provided on the request is
forwarded to the server, and canceling the context will make the function return at once.
* `func (c *Connection) NewStream(ctx context.Context, h HandlerID, payload []byte) (st *Stream, err error)`
will initiate a remote call and send the initial payload.
```Go
// A Stream is a two-way stream.
// All responses *must* be read by the caller.
// If the call is canceled through the context,
//The appropriate error will be returned.
type Stream struct {
// Responses from the remote server.
// Channel will be closed after an error or when the remote closes.
// All responses *must* be read by the caller until either an error is returned or the channel is closed.
// Canceling the context will cause the context cancellation error to be returned.
Responses <-chan Response
// Requests sent to the server.
// If the handler is defined with 0 incoming capacity this will be nil.
// Channel *must* be closed to signal the end of the stream.
// If the request context is canceled, the stream will no longer process requests.
Requests chan<- []byte
}
type Response struct {
Msg []byte
Err error
}
```
There are generic versions of the server/client handlers that allow the use of type
safe implementations for data types that support msgpack marshal/unmarshal.
sendfile implementation to perform DMA on all platforms
Go stdlib already supports sendfile/splice implementations
for
- Linux
- Windows
- *BSD
- Solaris
Along with this change however O_DIRECT for reads() must be
removed as well since we need to use sendfile() implementation
The main reason to add O_DIRECT for reads was to reduce the
chances of page-cache causing OOMs for MinIO, however it would
seem that avoiding buffer copies from user-space to kernel space
this issue is not a problem anymore.
There is no Go based memory allocation required, and neither
the page-cache is referenced back to MinIO. This page-
cache reference is fully owned by kernel at this point, this
essentially should solve the problem of page-cache build up.
With this now we also support SG - when NIC supports Scatter/Gather
https://en.wikipedia.org/wiki/Gather/scatter_(vector_addressing)
Tiering statistics have been broken for some time now, a regression
was introduced in 6f2406b0b6
Bonus fixes an issue where the objects are not assumed to be
of the 'STANDARD' storage-class for the objects that have
not yet tiered, this should be conditional based on the object's
metadata not a default assumption.
This PR also does some cleanup in terms of implementation,
fixes#18070
Disk level O_DIRECT support checking at xl storage initialization was
conditional on a config setting being enabled. (This never took effect
because config initialization happens after ObjectLayer is ready.) This
is not necessary as the config setting is dynamic - O_DIRECT should be
enabled via runtime config. So we need to do the disk level support
check regardless of the config setting.
not checking w.Close() can prematurely make us
think that the w.Write() actually succeeded, apparently
Write() may or may not return an error but sometimes
only during a Close() call to the fd we may see the
error from Write() propagate.
Fdatasync(w) on the FD would return an error requiring
Close() error handling is less of a concern, however it may
happen such that fdatasync() did not return an error, where
as Close() would.
Currently we have IOPs of these patterns
```
[OS] os.Mkdir play.min.io:9000 /disk1 2.718µs
[OS] os.Mkdir play.min.io:9000 /disk1/data 2.406µs
[OS] os.Mkdir play.min.io:9000 /disk1/data/.minio.sys 4.068µs
[OS] os.Mkdir play.min.io:9000 /disk1/data/.minio.sys/tmp 2.843µs
[OS] os.Mkdir play.min.io:9000 /disk1/data/.minio.sys/tmp/d89c8ceb-f8d1-4cc6-b483-280f87c4719f 20.152µs
```
It can be seen that we can save quite Nx levels such as
if your drive is mounted at `/disk1/minio` you can simply
skip sending an `Mkdir /disk1/` and `Mkdir /disk1/minio`.
Since they are expected to exist already, this PR adds a way
for us to ignore all paths upto the mount or a directory which
ever has been provided to MinIO setup.
objects with 10,000 parts and many of them can
cause a large memory spike which can potentially
lead to OOM due to lack of GC.
with previous PR reducing the memory usage significantly
in #17963, this PR reduces this further by 80% under
repeated calls.
Scanner sub-system has no use for the slice of Parts(),
it is better left empty.
```
benchmark old ns/op new ns/op delta
BenchmarkToFileInfo/ToFileInfo-8 295658 188143 -36.36%
benchmark old allocs new allocs delta
BenchmarkToFileInfo/ToFileInfo-8 61 60 -1.64%
benchmark old bytes new bytes delta
BenchmarkToFileInfo/ToFileInfo-8 1097210 227255 -79.29%
```
to track the replication transfer rate across different nodes,
number of active workers in use and in-queue stats to get
an idea of the current workload.
This PR also adds replication metrics to the site replication
status API. For site replication, prometheus metrics are
no longer at the bucket level - but at the cluster level.
Add prometheus metric to track credential errors since uptime
.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.
Bonus fixes include
- do not have to write final xl.meta (renameData) does this
already, saves some IOPs.
- make sure to purge the multipart directory properly using
a recursive delete, otherwise this can easily pile up and
rely on the stale uploads cleanup.
fixes#17863
slower drives get knocked off because they are too slow via
active monitoring, we do not need to block calls arbitrarily.
Serializing adds latencies for already slow calls, remove
it for SSDs/NVMEs
Also, add a selection with context when writing to `out <-`
channel, to avoid any potential blocks.
Removes the bloom filter since it has so limited usability, often gets saturated anyway and adds a bunch of complexity to the scanner.
Also removes a tiny bit of CPU by each write operation.
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
inlined data often is bigger than the allowed
O_DIRECT alignment, so potentially we can write
'xl.meta' without O_DSYNC instead we can rely on
O_DIRECT + fdatasync() instead.
This PR allows O_DIRECT on inlined data that
would gain the benefits of performing O_DIRECT,
eventually performing an fdatasync() at the end.
Performance boost can be observed here for small
objects < 128KiB. The performance boost is mainly
seen on HDD, and marginal on NVMe setups.
a/b/c/d/ where `a/b/c/` exists results in additional syscalls
such as an Lstat() call to verify if the `a/b/c/` exists
and its a directory.
We do not need to do this on MinIO since the parent prefixes
if exist, we can simply return success without spending
additional syscalls.
Also this implementation attempts to simply use Access() calls
to avoid os.Stat() calls since the latter does memory allocation
for things we do not need to use.
Access() is simpler since we have a predictable structure on
the backend and we know exactly how our path structures are.
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>
fix: change timedvalue to return previous cached value
caller can interpret the underlying error and decide
accordingly, places where we do not interpret the
errors upon timedValue.Get() - we should simply use
the previously cached value instead of returning "empty".
Bonus: remove some unused code
We need to make sure if we cannot read bucket metadata
for some reason, and bucket metadata is not missing and
returning corrupted information we should panic such
handlers to disallow I/O to protect the overall state
on the system.
In-case of such corruption we have a mechanism now
to force recreate the metadata on the bucket, using
`x-minio-force-create` header with `PUT /bucket` API
call.
Additionally fix the versioning config updated state
to be set properly for the site replication healing
to trigger correctly.
The test expects from DeleteFile to return errDiskNotFound when the disk
is not available. It calls os.RemoveAll() to remove one disk after XL storage
initialization. However, this latter contains some goroutines which can
race with os.RemoveAll() and then the test fails sporadically with
returning random errors.
The commit will tweak the initialization routine of the XL storage to
only run deletion of temporary and metacache data in the background,
so TestXLStorageDeleteFile won't fail anymore.
heal bucket metadata and IAM entries for
sites participating in site replication from
the site with the most updated entry.
Co-authored-by: Harshavardhana <harsha@minio.io>
Co-authored-by: Aditya Manthramurthy <aditya@minio.io>
This PR fixes two issues
- The first fix is a regression from #14555, the fix itself in #14555
is correct but the interpretation of that information by the
object layer code for "replication" was not correct. This PR
tries to fix this situation by making sure the "Delete" replication
works as expected when "VersionPurgeStatus" is already set.
Without this fix, there is a DELETE marker created incorrectly on
the source where the "DELETE" was triggered.
- The second fix is perhaps an older problem started since we inlined-data
on the disk for small objects, CopyObject() incorrectly inline's
a non-inlined data. This is due to the fact that we have code where
we read the `part.1` under certain conditions where the size of the
`part.1` is less than the specific "threshold".
This eventually causes problems when we are "deleting" the data that
is only inlined, which means dataDir is ignored leaving such
dataDir on the disk, that looks like an inconsistent content on
the namespace.
fixes#14767
```
tmp = buf[want:]
```
Would potentially crash when `buf` is truncated for some reason
and does not have the expected bytes, this is of course considered
not normal and is an odd situation. But we do not need to crash
here instead allow for errors to be returned and let callers handle
the errors.
This PR simply adds a warning message when it detects older kernel
versions and warn's them about potential performance issues on this
kernel.
The issue can be seen only with parallel I/O across all drives
on denser setups such as 90 drives or 45 drives per server configurations.
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.
This is a side-affect of the optimization done in PR #13544 which
causes a certain type of delete operations on given object versions
can cause lastVersion indication to be skipped, which leads to
an `xl.meta` where Versions[] slice is empty while the entire
file is intact by itself.
This PR tries to ensure that such files are visible and deletable
by regular means of listing as null 'delete-marker' and also
avoid the situation where this potential issue might arise.
- speedtest logs calls that were canceled
spuriously, in situations where it should
be ignored.
- all errors of interest are always sent back
to the client there is no need to log them
on the server console.
- PUT failures should negate the increments
such that GET is not attempted on unsuccessful
calls.
- do not attempt MRF on speedtest objects.
In the testing mode, reformatting disks will fail because the healing
code will complain if one disk is in root mode. This commit will
automatically set all disks as non-root if MINIO_CI_CD is set.
The current code considers a pool with all root disks to be as part
of a testing environment even if there are other pools with mounted
disks. This will result to illegitimate writing in root disks.
Fix this by simplifing the logic: require MINIO_CI_CD in order to skip
root disk check.
startup speed-up, currently getFormatErasureInQuorum()
would spend up to 2-3secs when there are 3000+ drives
for example in a setup, simplify this implementation
to use drive counts.
