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)
currently the default for all drives is 512, which is a lot
for HDDs the recent testing has revealed moving this to 32
for HDDs seems like a fair value.
.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:
- avoid calling DiskInfo() calls when missing blocks
instead heal the object using MRF operation.
- change the max_sleep to 250ms beyond that we will
not stop healing.
health checks were missing for drives replaced since
- HealFormat() would replace the drives without a health check
- disconnected drives when they reconnect via connectEndpoint()
the loop also loses health checks for local disks and merges
these into a single code.
- other than this separate cleanUp, health check variables to avoid
overloading them with similar requirements.
- also ensure that we compete via context selector for disk monitoring
such that the canceled disks don't linger around longer waiting for
the ticker to trigger.
- allow disabling active monitoring.
Add check every 2 minutes to see if a write+read operation can complete.
If disk is unresponsive for 2 minutes or returns errFaultyDisk, take it offline.
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>
Small uploads spend a significant amount of time (~5%) fetching disk info metrics. Also maps are allocated for each call.
Add a 100ms cache to disk metrics.
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.
Publish storage functions latency to help compare the performance
of different disks in a single deployment.
e.g.:
```
minio_node_disk_latency_us{api="storage.WalkDir",disk="/tmp/xl/1",server="localhost:9001"} 226
minio_node_disk_latency_us{api="storage.WalkDir",disk="/tmp/xl/2",server="localhost:9002"} 1180
minio_node_disk_latency_us{api="storage.WalkDir",disk="/tmp/xl/3",server="localhost:9003"} 1183
minio_node_disk_latency_us{api="storage.WalkDir",disk="/tmp/xl/4",server="localhost:9004"} 1625
```
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.
