This improves the startup time significantly
for clusters which have lot of buckets.
Also fixes a bug where `.minio.sys` is created
on disks which do not have `format.json`
startOffset was re-assigned to '0' so it would end up
copying wrong content ignoring the requested startOffset.
This also fixes the corruption issue we observed while
using docker registry.
Fixes https://github.com/docker/distribution/issues/2205
Also fixes#3842 - incorrect routing.
The globalMaxObjectSize limit is instilled in S3 spec perhaps
due to certain limitations on S3 infrastructure. For minio we
don't have such limitations and we can stream a larger file
instead.
So we are going to bump this limit to 16GiB.
Fixes#3825
This function was returning BucketNotFound for all errors
which at least hides the fact that disks could be corrupted.
This commit fixes the behavior by returning all errors that,
are, by the way, Object API errors.
Add missing protection from deleting multiple objects
in parallel. Currently we are deleting objects without
proper locking through this API.
This can cause significant amount of races.
Ignore a disk which wasn't able to successfully perform an action to
avoid eventual perturbations when the disk comes back in the middle
of write change.
This removal comes to avoid some redundant requirements
which are adding more problems on a production setup.
Here are the list of checks for time as they happen
- Fresh connect (during server startup) - CORRECT
- A reconnect after network disconnect - CORRECT
- For each RPC call - INCORRECT.
Verifying time for each RPC aggravates a situation
where a RPC call is rejected in a sequence of events
due to enough load on a production setup. 3 second
might not be enough time window for the call to be
initiated and received by the server.
Currently we document as IP:PORT which doesn't provide
if someone can use HOSTNAME:PORT. This is a change
to clarify this by calling it as ADDRESS:PORT which
encompasses both a HOSTNAME and an IP.
Fixes#3799
This PR is for readability cleanup
- getOrderedDisks as shuffleDisks
- getOrderedPartsMetadata as shufflePartsMetadata
Distribution is now a second argument instead being the
primary input argument for brevity.
Also change the usage of type casted int64(0), instead
rely on direct type reference as `var variable int64` everywhere.
Existing objects before overwrites are renamed to
temp location in completeMultipart. We make sure
that we delete it even if subsequenty calls fail.
Additionally move verifying of parent dir is a
file earlier to fail the entire operation.
Ref #3784
Content-Encoding is set to "aws-chunked" which is an S3 specific
API value which is no meaning for an object. This is how S3
behaves as well for a streaming signature uploaded object.
Make sure to skip reserved bucket names in `ListBuckets()`
current code didn't skip this properly and also generalize
this behavior for both XL and FS.
This is an attempt cleanup code and keep the top level config
functions simpler and easy to understand where as move the
notifier related code and logger setter/getter methods as part
of their own struct.
Locks are now held properly not globally by configMutex, but
instead as private variables.
Final fix for #3700
Also changes the behavior of `secretKeyHash` which is
not necessary to be sent over the network, each node
has its own secretKeyHash to validate.
Fixes#3696
Partial(fix) #3700 (More changes needed with some code cleanup)
Currently the auth rpc client defaults to to a maximum
cap of 30seconds timeout. Make this to be configurable
by the caller of authRPCClient during initialization, if no
such config is provided then default to 30 seconds.
Ideally here if the interface is not found it would
fail the server, as it should be because without these
we can't even have a working server in the first place.
Just like how it fails in master invariably inside Go
net/http code path.
Fixes#3708
Network: total bytes of incoming and outgoing server's data
by taking advantage of our ConnMux Read/Write wrapping
HTTP: total number of different http verbs passed in http
requests and different status codes passed in http responses.
This is counted in a new http handler.
Resource strings and paths are case insensitive on windows
deployments but if user happens to use upper case instead of
lower case for certain configuration params like bucket
policies and bucket notification config. We might not honor
them which leads to a wrong behavior on windows.
This is windows only behavior, for all other platforms case
is still kept sensitive.
Avoid passing size = -1 to PutObject API by requiring content-length
header in POST request (as AWS S3 does) and in Upload web handler.
Post handler is modified to completely store multipart file to know
its size before sending it to PutObject().
Following is a sample list lock API request schematic,
/?lock&bucket=mybucket&prefix=myprefix&duration=holdDuration
x-minio-operation: list
The response would contain the list of locks held on mybucket matching
myprefix for a duration longer than holdDuration.
Current implementation didn't honor quorum properly and didn't
handle the errors generated properly. This patch addresses that
and also moves common code `cleanupMultipartUploads` into xl
specific private function.
Fixes#3665
On macOS, if a process already listens on 127.0.0.1:PORT, net.Listen() falls back
to IPv6 address ie minio will start listening on IPv6 address whereas another
(non-)minio process is listening on IPv4 of given port.
To avoid this error sutiation we check for port availability only for macOS.
Note: checkPortAvailability() tries to listen on given port and closes it.
It is possible to have a disconnected client in this tiny window of time.
Creds don't require secretKeyHash to be calculated
everytime, cache it instead and re-use.
This is an optimization for bcrypt.
Relevant results from the benchmark done locally, negative
value means improvement in this scenario.
```
benchmark old ns/op new ns/op delta
BenchmarkAuthenticateNode-4 160590992 80125647 -50.11%
BenchmarkAuthenticateWeb-4 160556692 80432144 -49.90%
benchmark old allocs new allocs delta
BenchmarkAuthenticateNode-4 87 75 -13.79%
BenchmarkAuthenticateWeb-4 87 75 -13.79%
benchmark old bytes new bytes delta
BenchmarkAuthenticateNode-4 15222 9785 -35.72%
BenchmarkAuthenticateWeb-4 15222 9785 -35.72%
```
An external test that runs cmd.Main() has a difficulty to set cmd arguments
and MINIO_{ACCESS,SECRET}_KEY values, this commit changes a little the current
behavior in a way that helps external tests.
Encode the path of the passed presigned url before calculating the signature. This fixes
presigning objects whose names contain characters that are found encoded in urls.
* Implement heal format REST API handler
* Implement admin peer rpc handler to re-initialize storage
* Implement HealFormat API in pkg/madmin
* Update pkg/madmin API.md to incl. HealFormat
* Added unit tests for ReInitDisks rpc handler and HealFormatHandler
For TLS peekProtocol do not assume the incoming request to be a TLS
connection perform a handshake() instead and validate.
Also add some security related defaults to `tls.Config`.
This restriction has lots of side affects, since
we do not have a mechanism to clear states like
this it is better not to keep them.
Network errors are common and can occur with
simple cable removal etc. Since we already have
a retry mechanism this error count and stateful
nature can bring problems on a long running
cluster.
This is a consolidation effort, avoiding usage
of naked strings in codebase. Whenever possible
use constants which can be repurposed elsewhere.
This also fixes `goconst ./...` reported issues.
`principalId` i.e user identity is kept as AccessKey in
accordance with S3 spec.
Additionally responseElements{} are added starting with
`x-amz-request-id` is a hexadecimal of the event time itself in nanosecs.
`x-minio-origin-server` - points to the server generating the event.
Fixes#3556
URL paths can be empty and not have preceding separator,
we do not yet know the conditions this can happen inside
Go http server.
This patch is to ensure that we do not crash ourselves
under conditions where r.URL.Path may be empty.
Fixes#3553
A client sends escaped characters in values of some query parameters in a presign url.
This commit properly unescapes queires to fix signature calculation.
Golang HTTP client automatically detects content-type but
for S3 clients this content-type might be incorrect or
might misbehave.
For example:
```
Content-Type: text/xml; charset=utf-8
```
Should be
```
Content-Type: application/xml
```
Allow this to be set properly.
* Filter lock info based on bucket, prefix and time since lock was held
* Implement list and clear locks REST API
* madmin: Add list and clear locks API
* locks: Clear locks matching bucket, prefix, relTime.
* Gather lock information across nodes for both list and clear locks admin REST API.
* docs: Add lock API to management APIs
* Rename GenericArgs to AuthRPCArgs
* Rename GenericReply to AuthRPCReply
* Remove authConfig.loginMethod and add authConfig.ServiceName
* Rename loginServer to AuthRPCServer
* Rename RPCLoginArgs to LoginRPCArgs
* Rename RPCLoginReply to LoginRPCReply
* Version and RequestTime are added to LoginRPCArgs and verified by
server side, not client side.
* Fix data race in lockMaintainence loop.
This patch uses a technique where in a retryable storage
before object layer initialization has a higher delay
and waits for longer period upto 4 times with time unit
of seconds.
And uses another set of configuration after the disks
have been formatted, i.e use a lower retry backoff rate
and retrying only once per 5 millisecond.
Network IO error count is reduced to a lower value i.e 256
before we reject the disk completely. This is done so that
combination of retry logic and total error count roughly
come to around 2.5secs which is when we basically take the
disk offline completely.
NOTE: This patch doesn't fix the issue of what if the disk
is completely dead and comes back again after the initialization.
Such a mutating state requires a change in our startup sequence
which will be done subsequently. This is an interim fix to alleviate
users from these issues.
Implement a storage rpc specific rpc client,
which does not reconnect unnecessarily.
Instead reconnect is handled at a different
layer for storage alone.
Rest of the calls using AuthRPC automatically
reconnect, i.e upon an error equal to `rpc.ErrShutdown`
they dial again and call the requested method again.
Attempt a reconnect also if disk not found.
This is needed since any network operation error
is converted to disk not found but we also need
to make sure if disk is really not available.
Additionally we also need to retry more than
once because the server might be in startup
sequence which would render other servers to
wrongly think that the server is offline.
This is written so that to simplify our handler code
and provide a way to only update metadata instead of
the data when source and destination in CopyObject
request are same.
Fixes#3316
- Add a lockStat type to group counters
- Remove unnecessary helper functions
- Fix stats computation on force unlock
- Removed unnecessary checks and cleaned up comments
This is to utilize an optimized version of
sha256 checksum which @fwessels implemented.
blake2b lacks such optimizations on ARM platform,
this can provide us significant boost in performance.
blake2b on ARM64 as expected would be slower.
```
BenchmarkSize1K-4 30000 44015 ns/op 23.26 MB/s
BenchmarkSize8K-4 5000 335448 ns/op 24.42 MB/s
BenchmarkSize32K-4 1000 1333960 ns/op 24.56 MB/s
BenchmarkSize128K-4 300 5328286 ns/op 24.60 MB/s
```
sha256 on ARM64 is faster by orders of magnitude giving close to
AVX performance of blake2b.
```
BenchmarkHash8Bytes-4 1000000 1446 ns/op 5.53 MB/s
BenchmarkHash1K-4 500000 3229 ns/op 317.12 MB/s
BenchmarkHash8K-4 100000 14430 ns/op 567.69 MB/s
BenchmarkHash1M-4 1000 1640126 ns/op 639.33 MB/s
```
ObjectLayer GetObject() now returns the entire object
if starting offset is 0 and length is negative. This
also allows to simplify handler layer code where
we always had to use GetObjectInfo() before proceeding
to read bucket metadata files examples `policy.json`.
This also reduces one additional call overhead.
success_action_redirect in the sent Form means that the server needs to return 303 in addition to a well specific redirection url, this commit adds this feature
This is important in a distributed setup, where the server hosting the
first disk formats a fresh setup. Sorting ensures that all servers
arrive at the same 'first' server.
Note: This change doesn't protect against different disk arguments
with some disks being same across servers.
Previously, more than one goroutine calls RPCClient.dial(), each
goroutine gets a new rpc.Client but only one such client is stored
into RPCClient object. This leads to leaky connection at the server
side. This is fixed by taking lock at top of dial() and release on
return.
There was an error in how we validated disk formats,
if one of the disk was formatted and was formatted with
FS would cause confusion and object layer would never
initialize essentially go into an infinite loop.
Validate pre-emptively and also check for FS format
properly.
This is implemented so that the issues like in the
following flow don't affect the behavior of operation.
```
GetObjectInfo()
.... --> Time window for mutation (no lock held)
.... --> Time window for mutation (no lock held)
GetObject()
```
This happens when two simultaneous uploads are made
to the same object the object has returned wrong
info to the client.
Another classic example is "CopyObject" API itself
which reads from a source object and copies to
destination object.
Fixes#3370Fixes#2912
FS/Multipart: Fix race between PutObjectPart and Complete/Abort multipart. close(timeoutCh) on complete/abort so that a racing PutObjectPart does not leave a dangling go-routine.
Fixes#3351
This change brings in changes at multiple places
- Reuse buffers at almost all locations ranging
from rpc, fs, xl, checksum etc.
- Change caching behavior to disable itself
under low memory conditions i.e < 8GB of RAM.
- Only objects cached are of size 1/10th the size
of the cache for example if 4GB is the cache size
the maximum object size which will be cached
is going to be 400MB. This change is an
optimization to cache more objects rather
than few larger objects.
- If object cache is enabled default GC
percent has been reduced to 20% in lieu
with newly found behavior of GC. If the cache
utilization reaches 75% of the maximum value
GC percent is reduced to 10% to make GC
more aggressive.
- Do not use *bytes.Buffer* due to its growth
requirements. For every allocation *bytes.Buffer*
allocates an additional buffer for its internal
purposes. This is undesirable for us, so
implemented a new cappedWriter which is capped to a
desired size, beyond this all writes rejected.
Possible fix for #3403.
- This is to ensure that the any new config references made to the
serverConfig is also backed by a mutex lock.
- Otherwise any new config assigment will also replace the member mutex
which is currently used for safe access.
EOF err message in Peek Protocol is shown when a client closes the
connection in the middle of peek protocol, this commit hides it since it
doesn't make sense to show it
Current code always appends to a file only if 1byte or
more was sent on the wire was affecting both PutObject
and PutObjectPart uploads.
This patch fixes such a situation and resolves#3385
backgroundAppend type's abort method should wait for appendParts to finish
writing ongoing appending of parts in the background before cleaning up
the part files.
Previously minio server expects content-length-range values as integer
in JSON. However Amazon S3 handles content-length-range values as
integer and strings.
This patch adds support for string values.
Since we moved out reconnection logic from net-rpc-client.go
we should do it from the top-layer properly and bring back
the code to reconnect properly in-case the connection is lost.
setGlobalsFromContext() is added to set global variables after parsing
command line arguments. Thus, global flags will be honored wherever
they are placed in minio command.
Make sure all S3 signature requests are not re-directed
to `/minio`. This should be only done for JWT and some
Anonymous requests.
This also fixes a bug found from https://github.com/bji/libs3
```
$ s3 -u list
ERROR: XmlParseFailure
```
Now after this fix shows proper output
```
$ s3 -u list
Bucket Created
-------------------------------------------------------- --------------------
andoria 2016-11-27T08:19:06Z
```
It would make sense to enable logger just after config initialisation.
That way, errorIf() and fatalIf() will be usable and can catch error
like invalid access and key errors.
This patch brings in changes from miniobrowser repo.
- Bucket policy UI and functionality fixes by @krishnasrinivas
- Bucket policy implementation by @balamurugana
- UI changes and new functionality changing password etc. @rushenn
- UI and new functionality for sharing URLs, deleting files
@rushenn and @krishnasrinivas.
- Other misc fixes by @vadmeste @brendanashworth
This is needed to validate if the `format.json` indeed exists
when a fresh node is brought online.
This wrapped implementation also connects to the remote node
by attempting a re-login. Subsequently after a successful
connect `format.json` is validated as well.
Fixes#3207
logurs is not helping us to set different log formats (json/text) to
different loggers. Now, we create different logurs instances and call
them in errorIf and fatalIf
This change adds more richer error response
for JSON-RPC by interpreting object layer
errors to corresponding meaningful errors
for the web browser.
```go
&json2.Error{
Message: "Bucket Name Invalid, Only lowercase letters, full stops, and numbers are allowed.",
}
```
Additionally this patch also allows PresignedGetObject()
to take expiry parameter to have variable expiry.
XL multipart fails to remove tmp files when an error occurs during upload, this case covers the scenario where an upload is canceled manually by the client in the middle of job.
content-length-range policy in postPolicy API was
not working properly handle it. The reflection
strategy used has changed in recent version of Go.
Any free form interface{} of any integer is treated
as `float64` this caused a bug where content-length-range
parsing failed to provide any value.
Fixes#3295
This is needed as explained by @krisis
Lets say we have following errors.
```
[]error{nil, errFileNotFound, errDiskAccessDenied, errDiskAccesDenied}
```
Since the last two errors are filtered, the maximum is nil,
depending on map order.
Let's say we get nil from reduceErr. Clearly at this point
we don't have quorum nodes agreeing about the data and since
GetObject only requires N/2 (Read quorum) and isDiskQuorum
would have returned true. This is problematic and can lead to
undersiable consequences.
Fixes#3298
Disks when are offline for a long period of time, we should
ignore the disk after trying Login upto 5 times.
This is to reduce the network chattiness, this also reduces
the overall time spent on `net.Dial`.
Fixes#3286
For binary releases and operating systems it would be
All operating systems.
```
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Minio is 25 days 12 hours 30 minutes old ┃
┃ Update: https://dl.minio.io/server/minio/release/linux-amd64/minio ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
```
On docker.
```
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Minio is 25 days 12 hours 32 minutes old ┃
┃ Update: docker pull minio/minio ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
```
In a situation when we have lots of buckets the bootup time
might have slowed down a bit but during this situation the
servers quickly going up and down would be an in-transit state.
Certain calls which do not use quorum like `readXLMetaStat`
might return an error saying `errDiskNotFound` this is returned
in place of expected `errFileNotFound` which leads to an issue
where server doesn't start.
To avoid this situation we need to ignore them as safe values
to be ignored, for the most part these are network related errors.
Fixes#3275
Also fix test to not use a bucket name with a leading slash - this
causes the bucket name to become empty and go to an unintended API
call (listbuckets).
This patch fixes a possible bug, reproduced rarely only seen
once.
```
panic: runtime error: index out of range
goroutine 136 [running]:
panic(0xac1a40, 0xc4200120b0)
/usr/local/go/src/runtime/panic.go:500 +0x1a1
github.com/minio/minio/vendor/github.com/minio/dsync.lock.func1(0xc4203d2240, 0x4, 0xc420474080, 0x4, 0x4, 0xc4202abb60, 0x0, 0xa86d01, 0xefcfc0, 0xc420417a80)
/go/src/github.com/minio/minio/vendor/github.com/minio/dsync/drwmutex.go:170 +0x69b
created by github.com/minio/minio/vendor/github.com/minio/dsync.lock
/go/src/github.com/minio/minio/vendor/github.com/minio/dsync/drwmutex.go:191 +0xf4
```
Ref #3229
After review with @abperiasamy we decided to remove all the unnecessary options
- MINIO_BROWSER (Implemented as a security feature but now deemed obsolete
since even if blocking access to MINIO_BROWSER, s3 API port is open)
- MINIO_CACHE_EXPIRY (Defaults to 72h)
- MINIO_MAXCONN (No one used this option and we don't test this)
- MINIO_ENABLE_FSMETA (Enable FSMETA all the time)
Remove --ignore-disks option - this option was implemented when XL layer
would initialize the backend disks and heal them automatically to disallow
XL accidentally using the root partition itself this option was introduced.
This behavior has been changed XL no longer automatically initializes
`format.json` a HEAL is controlled activity, so ignore-disks is not
useful anymore. This change also addresses the problems of our documentation
going forward and keeps things simple. This patch brings in reduction of
options and defaulting them to a valid known inputs. This patch also
serves as a guideline of limiting many ways to do the same thing.
rpcClient should attempt a reconnect if the call fails
with 'rpc.ErrShutdown' this is needed since at times when
the servers are taken down and brought back up.
The hijacked connection from net.Dial is usually closed.
So upon first attempt rpcClient might falsely indicate that
disk to be down, to avoid this state make another dial attempt
to really fail.
Fixes#3206Fixes#3205
- abstract out instrumentation information.
- use separate lockInstance type that encapsulates the nsMutex, volume,
path and opsID as the frontend or top-level lock object.
This is done by not making the methods of the BucketMetaState interface
as methods (via type nesting) on the type implementing
RPCs (s3PeerAPIHandlers).
- Adds an interface to update in-memory bucket metadata state called
BucketMetaState - this interface has functions to:
- update bucket notification configuration,
- bucket listener configuration,
- bucket policy configuration, and
- send bucket event
- This interface is implemented by `localBMS` a type for manipulating
local node in-memory bucket metadata, and by `remoteBMS` a type for
manipulating remote node in-memory bucket metadata.
- The remote node interface, makes an RPC call, but the local node
interface does not - it updates in-memory bucket state directly.
- Rename mkPeersFromEndpoints to makeS3Peers and refactored it.
- Use arrayslice instead of map in s3Peers struct
- `s3Peers.SendUpdate` now receives an arrayslice of peer indexes to
send the request to, with a special nil value slice indicating that
all peers should be sent the update.
- `s3Peers.SendUpdate` now returns an arrayslice of errors, representing
errors from peers when sending an update. The array positions
correspond to peer array s3Peers.peers
Improve globalS3Peers:
- Make isDistXL a global `globalIsDistXL` and remove from s3Peers
- Make globalS3Peers an array of (address, bucket-meta-state) pairs.
- Fix code and tests.
Default golang net.Listen only listens on the first IP when
host resolves to multiple IPs.
This change addresses a problem for example your ``/etc/hosts``
has entries as following
```
127.0.1.1 minio1
192.168.1.10 minio1
```
Trying to start minio as
```
minio server --address "minio1:9001" ~/Photos
```
Causes the minio server to be bound only to "127.0.1.1" which
is an incorrect behavior since we are generally interested in
`192.168.1.10` as well.
This patch addresses this issue if the hostname is resolvable
and gives back list of addresses associated with that hostname
we just bind on all of them as it is the expected behavior.
- The benchmark initialization function was not taking into account the
instance type (FS/XL), was using XL ObjectLayer even for FS
benchmarks.
- This was leading to incorrect benchmark results for FS related
benchmarks.
- The fix takes into account the instance type (FS/XL) and correctly
returns FS backend for FS benchmarks.
