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.
With an odd number of drives per erasure set setup, the write/quorum is
the half + 1; however the decommissioning listing will still list those
objects and does not consider those as stale.
Fix it by using (N+1)/2 formula.
Co-authored-by: Anis Elleuch <anis@min.io>
Immediate transition use case and is mostly used to fill warm
backend with a lot of data when a new deployment is created
Currently, if the transition queue is complete, the transition will be
deferred to the scanner; change this behavior by blocking the PUT request
until the transition queue has a new place for a transition task.
Currently if the object does not exist in quorum disks of an erasure
set, the dangling code is never called because the returned error will
be errFileNotFound or errFileVersionNotFound;
With this commit, when errFileNotFound or errFileVersionNotFound is
returning when trying to calculate the quorum of a given object, the
code checks if a disk returned nil, which means a stale object exists in
that disk, that will trigger deleteIfDangling() function
This commit splits the liveness and readiness
handler into two separate handlers. In K8S, a
liveness probe is used to determine whether the
pod is in "live" state and functioning at all.
In contrast, the readiness probe is used to
determine whether the pod is ready to serve
requests.
A failing liveness probe causes pod restarts while
a failing readiness probe causes k8s to stop routing
traffic to the pod. Hence, a liveness probe should
be as robust as possible while a readiness probe
should be used to load balancing.
Ref: https://kubernetes.io/docs/tasks/configure-pod-container/configure-liveness-readiness-startup-probes/
Signed-off-by: Andreas Auernhammer <github@aead.dev>
This patch takes care of loading the bucket configs of failed buckets
during the periodic refresh. This makes sure the event notifiers and
remote bucket targets are properly initialized.
users might use MinIO on NFS, GPFS that provide dynamic
inodes and may not even have a concept of free inodes.
to allow users to use MinIO on top of GPFS relax the
free inode check.
* creating a byte buffer for SFTP file segments
* Adding an error condition for when there are
remaining segments in the queue
* Simplification of the queue using a map
it is possible that ILM or Deletes got triggered on batch
of objects that we are attempting to batch replicate, ignore
this scenario as valid behavior.
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)
`monitorAndConnectEndpoints` will continue to attempt to reconnect offline disks.
Since disks were never closed, a `MarkOffline` would continue to try to check these disks forever.
Close previous disks.
replace io.Discard usage to fix NUMA copy() latencies
On NUMA systems copying from 8K buffer allocated via
io.Discard leads to large latency build-up for every
```
copy(new8kbuf, largebuf)
```
can in-cur upto 1ms worth of latencies on NUMA systems
due to memory sharding across NUMA nodes.
Fix various regressions from #18029
* If context is canceled the token is never returned. This will lead to scanner being unable to save and deadlocking.
* Fix backup not being able to get any data (hr empty)
* Reduce backup timeout.
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
https://github.com/minio/minio/pull/18307 partially removed the duplicate upload id check.
While I can't really see how ListDir can return duplicate entries, let's re-add it, since it is a cheap sanity check.
There can be rare situations where errors seen in bucket metadata
load on startup or subsequent metadata updates can result in missing
replication remotes.
Attempt a refresh of remote targets backed by a good replication config
lazily in 5 minute intervals if there ever occurs a situation where
remote targets go AWOL.
Since relaxing quorum the error across pools
for ListBuckets(), GetBucketInfo() we hit a
situation where loading IAM could potentially
return an error for second pool that server
is not initialized.
We need to handle this, let the pool come online
and retry transparently - this PR fixes that.
x-amz-signed-headers is meant for HTTP headers only
not for query params, using that to verify things
further can lead to failure.
The generated presigned URL with custom metadata
is already kosher (tamper proof).
fixes#18281
`resourceMetricsMap` has no protection against concurrent reads and writes.
Add a mutex and don't use maps from the last iteration.
Bug introduced in #18057Fixes#18271
globalDeploymentID was being read while it was being set.
Fixes race:
```
WARNING: DATA RACE
Write at 0x0000079605a0 by main goroutine:
github.com/minio/minio/cmd.connectLoadInitFormats()
github.com/minio/minio/cmd/prepare-storage.go:269 +0x14f0
github.com/minio/minio/cmd.waitForFormatErasure()
github.com/minio/minio/cmd/prepare-storage.go:294 +0x21d
...
Previous read at 0x0000079605a0 by goroutine 105:
github.com/minio/minio/cmd.newContext()
github.com/minio/minio/cmd/utils.go:817 +0x31e
github.com/minio/minio/cmd.adminMiddleware.func1()
github.com/minio/minio/cmd/admin-router.go:110 +0x96
net/http.HandlerFunc.ServeHTTP()
net/http/server.go:2136 +0x47
github.com/minio/minio/cmd.setBucketForwardingMiddleware.func1()
github.com/minio/minio/cmd/generic-handlers.go:460 +0xb1a
net/http.HandlerFunc.ServeHTTP()
net/http/server.go:2136 +0x47
...
```
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.
Introducing a new version of healthinfo struct for adding this info is
not correct. It needs to be implemented differently without adding a new
version.
This reverts commit 8737025d940f80360ed4b3686b332db5156f6659.
There is a fundamental race condition in `newErasureServerPools`, where setObjectLayer is
called before the poolMeta has been loaded/populated.
We add a placeholder value to this field but disable all saving of the value, so we don't risk
overwriting the value on disk. Once the value has been loaded or created, it is replaced with
the proper value, which will also be saved.
Also fixes various accesses of `poolMeta` that were done without locks.
We make the `poolMeta.IsSuspended` return false, even if we shouldn't risk out-of-bounds
reads anymore.
if erasure upgrade is needed rely on the in-memory
values, instead of performing a "DiskInfo()" call.
https://brendangregg.com/blog/2016-09-03/sudden-disk-busy.html
for HDDs these are problematic, lets avoid this because
there is no value in "being" absolutely strict here
in terms of parity. We are okay to increase parity
as we see based on the in-memory online/offline ratio.
Several callers to putObjectTar may be fighting to set sc. Move the write out of the loop.
Use static resp, and request elements.
Fixes tests with -race:
```
WARNING: DATA RACE
Read at 0x00c01cd680e0 by goroutine 691354:
github.com/minio/minio/cmd.objectAPIHandlers.PutObjectExtractHandler.func1()
e:/gopath/src/github.com/minio/minio/cmd/object-handlers.go:2130 +0x149
github.com/minio/minio/cmd.untar.func1()
e:/gopath/src/github.com/minio/minio/cmd/untar.go:250 +0x2b6
github.com/minio/minio/cmd.untar.func8()
e:/gopath/src/github.com/minio/minio/cmd/untar.go:261 +0xa4
Previous write at 0x00c01cd680e0 by goroutine 691352:
github.com/minio/minio/cmd.objectAPIHandlers.PutObjectExtractHandler.func1()
e:/gopath/src/github.com/minio/minio/cmd/object-handlers.go:2131 +0x15d
github.com/minio/minio/cmd.untar.func1()
e:/gopath/src/github.com/minio/minio/cmd/untar.go:250 +0x2b6
github.com/minio/minio/cmd.untar.func8()
e:/gopath/src/github.com/minio/minio/cmd/untar.go:261 +0xa4
```
Calling unfreezeServices twice results in panic:
```
panic: "POST /minio/peer/v32/signalservice?signal=4&sub-sys=": close of nil channel
goroutine 14703 [running]:
runtime/debug.Stack()
runtime/debug/stack.go:24 +0x65
github.com/minio/minio/cmd.setCriticalErrorHandler.func1.1()
github.com/minio/minio/cmd/generic-handlers.go:549 +0x8e
panic({0x27c3020, 0x4c9b370})
runtime/panic.go:884 +0x212
github.com/minio/minio/cmd.unfreezeServices()
github.com/minio/minio/cmd/service.go:112 +0xc7
github.com/minio/minio/cmd.(*peerRESTServer).SignalServiceHandler(0x0?, {0x4cb6af0, 0xc010b96420}, 0xc01affab00)
github.com/minio/minio/cmd/peer-rest-server.go:837 +0x13a
net/http.HandlerFunc.ServeHTTP(...)
```
If the function was called a second time `val` would not be nil, but the returned channel `ch` would be, causing the panic.
Check the channel isn't nil and also use Swap for an atomic swap instead of 2 separate operations (though we are in a mutex).
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.
- Trace needs higher buffered channels than 4000 to ensure
when we run `mc admin trace -a` it captures all information
sufficiently.
- Listen event notification needs the event channel to be
`apiRequestsMaxPerNode` * number of nodes
Currently, the retry is not fully used when there is no backup copy of
the data usage; use 5 retry attempts when we don't have any valid data,
new or backup, unless we have seen an un-recognized error.
comment in the code provides more detailed explanation
on what this PR entails and its assumptions.
this PR reduces the amount of listing() by an order
of magnitude, however there are other such calls that
still needs further optimization that shall be done
in subsequent PRs.
Add a new endpoint for "resource" metrics `/v2/metrics/resource`
This should return system metrics related to drives, network, CPU and
memory. Except for drives, other metrics should have corresponding "avg"
and "max" values also.
Reuse the real-time feature to capture the required data,
introducing CPU and memory metrics in it.
Collect the data every minute and keep updating the average and max values
accordingly, returning the latest values when the API is called.
without this the rename2() can rename the previous dataDir
causing issues for different versions of the object, only
latest version is preserved due to this bug.
Added healing code to ensure recovery of such content.
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, setting a new tiering target returns an error when a bucket
is versioned and the tiering credentials does not have authorization to
specify a version-id when reading or removing a specific version;
Since tiering does not require versioning anymore; avoid doing versioned
operations when performing checklist ops while adding a new tiering
configuration.
Do not error out when a provided marker is before or after the prefix, but instead just ignore it if before and return an empty list when after.
Fixes#18093
Include object and versions heal scan times when checking non-empty abandoned folders.
Furthermore don't add delay between healing versions, instead do one per object wait.
This PR changes the StatObject() to be must have for non-minio source
to being a conditional API call.
- Calls StatObject() when needed
- Calls GetObjectTagging() when needed
These calls if we do without these conditionals can cause a lot of
delays, so we avoid them if not needed in more common scenario.
If MinIO started with KMS enabled, MINIO_KMS_KES_KEY_NAME should
be set for server to start.
Signed-off-by: Shubhendu Ram Tripathi <shubhendu@minio.io>
In a perf test, one node will run speed test with all nodes. If there is
an error with a peer node, the peer node name is not included in the
error hence confusing the user.
This commit will add the peer endpoint string to the netperf error.
To ensure that policy mappings are current for service accounts
belonging to (non-derived) STS accounts (like an LDAP user's service
account) we periodically reload such mappings.
This is primarily to handle a case where a policy mapping update
notification is missed by a minio node. Such a node would continue to
have the stale mapping in memory because STS creds/mappings were never
periodically scanned from storage.
- we already have MRF for most recent failures
- we trigger healing during HEAD/GET operation
These are enough, also change the default max wait
from 5sec to 1sec for default scanner speed.
AccountInfo is quite frequently called by the Console UI
login attempts, when many users are logging in it is important
that we provide them with better responsiveness.
- ListBuckets information is cached every second
- Bucket usage info is cached for up to 10 seconds
- Prefix usage (optional) info is cached for up to 10 secs
Failure to update after cache expiration, would still
allow login which would end up providing information
previously cached.
This allows for seamless responsiveness for the Console UI
logins, and overall responsiveness on a heavily loaded
system.
From the Go specification:
"3. If the map is nil, the number of iterations is 0." [1]
Therefore, an additional nil check for before the loop is unnecessary.
[1]: https://go.dev/ref/spec#For_range
Signed-off-by: Eng Zer Jun <engzerjun@gmail.com>
- remove targetClient for passing around via replicationObjectInfo{}
- remove cloing to object info unnecessarily
- remove objectInfo from replicationObjectInfo{} (only require necessary fields)
When using a chain provider all providers do not return a valid
access and secret key, an anonymous request is sent, which makes it hard
for users to figure out what is going on
In the case of S3 tiering, when AWS IAM temporary account generation returns
an error, an anonymous login will be used because of the chain provider.
Avoid this and use the AWS IAM provider directly to get a good error
message.
This helps reduce disk operations as these periodic routines would not
run concurrently any more.
Also add expired STS purging periodic operation: Since we do not scan
the on-disk STS credentials (and instead only load them on-demand) a
separate routine is needed to purge expired credentials from storage.
Currently this runs about a quarter as often as IAM refresh.
Also fix a bug where with etcd, STS accounts could get loaded into the
iamUsersMap instead of the iamSTSAccountsMap.
This allows scanner to avoid lengthy scans, skip
things appropriately and also not lose metrics in
any manner.
reduce longer deadlines for usage-cache loads/saves
to match the disk timeout which is 2minutes now per
IOP.
In situations with large number of STS credentials on disk, IAM load
time is high. To mitigate this, STS accounts will now be loaded into
memory only on demand - i.e. when the credential is used.
In each IAM cache (re)load we skip loading STS credentials and STS
policy mappings into memory. Since STS accounts only expire and cannot
be deleted, there is no risk of invalid credentials being reused,
because credential validity is checked when it is used.
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.
Previously existing objects were queued to single worker and MRF re-queues
are also handled by same worker - this does not fully use the available
bandwidth in case there is no incoming workload.
Errors such as
```
returned an error (context deadline exceeded) (*fmt.wrapError)
```
```
(msgp: too few bytes left to read object) (*fmt.wrapError)
```
This change enables embedding files in ZIP with custom permissions.
Also uses default creds for starting MinIO based on inspect data.
Signed-off-by: Shubhendu Ram Tripathi <shubhendu@minio.io>
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%
```
- this PR avoids sending a large ChecksumInfo slice
when its not needed
- also for a file with XLV2 format there is no reason
to allocate Checksum slice while reading
Keys are helpful to ensure the strict ordering of messages, however currently the
code uses a random request id for every log, hence using the request-id
as a Kafka key is not serve any purpose;
This commit removes the usage of the key, to also fix the audit issue from
internal subsystem that does not have a request ID.
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
replicationTimestamp might differ if there were retries
in replication and the retried attempt overwrote in
quorum but enough shards with newer timestamp causing
the existing timestamps on xl.meta to be invalid, we
do not rely on this value for anything external.
this is purely a hint for debugging purposes, but there
is no real value in it considering the object itself
is in-tact we do not have to spend time healing this
situation.
we may consider healing this situation in future but
that needs to be decoupled to make sure that we do not
over calculate how much we have to heal.
.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.
given a versionId the mtime is always the same, it
can never be different than its original value.
versionIds also do not conflict, since they are uuid's
and unique practically forever.
we expect a certain level of IOPs and latency so this is okay.
fixes other miscellaneous bugs
- such as hanging on mrfCh <- when the context is canceled
- queuing MRF heal when the context is canceled
- remove unused saveStateCh channel
Krishnan Parthasarathi
Klaus Post
Anis Eleuch
Sveinn
Harshavardhana
Shubhendu
Andreas Auernhammer
Praveen raj Mani
Taran Pelkey
Poorna
jiuker
vicmunoz
Shireesh Anjal
Daniel López Guimaraes
Allan Roger Reid
Adrian Najera
Matthew Toohey
Aditya Manthramurthy
Maxim Tkachenko
Eng Zer Jun
Alex
yangw
Daniel Valdivia