major performance improvements in range GETs to avoid large
read amplification when ranges are tiny and random
```
-------------------
Operation: GET
Operations: 142014 -> 339421
Duration: 4m50s -> 4m56s
* Average: +139.41% (+1177.3 MiB/s) throughput, +139.11% (+658.4) obj/s
* Fastest: +125.24% (+1207.4 MiB/s) throughput, +132.32% (+612.9) obj/s
* 50% Median: +139.06% (+1175.7 MiB/s) throughput, +133.46% (+660.9) obj/s
* Slowest: +203.40% (+1267.9 MiB/s) throughput, +198.59% (+753.5) obj/s
```
TTFB from 10MiB BlockSize
```
* First Access TTFB: Avg: 81ms, Median: 61ms, Best: 20ms, Worst: 2.056s
```
TTFB from 1MiB BlockSize
```
* First Access TTFB: Avg: 22ms, Median: 21ms, Best: 8ms, Worst: 91ms
```
Full object reads however do see a slight change which won't be
noticeable in real world, so not doing any comparisons
TTFB still had improvements with full object reads with 1MiB
```
* First Access TTFB: Avg: 68ms, Median: 35ms, Best: 11ms, Worst: 1.16s
```
v/s
TTFB with 10MiB
```
* First Access TTFB: Avg: 388ms, Median: 98ms, Best: 20ms, Worst: 4.156s
```
This change should affect all new uploads, previous uploads should
continue to work with business as usual. But dramatic improvements can
be seen with these changes.
This commit adds the `FromContentMD5` function to
parse a client-provided content-md5 as ETag.
Further, it also adds multipart ETag computation
for future needs.
The base profiles contains no valuable data, don't record them.
Reduce block rate by 2 orders of magnitude, should still capture just as valuable data with less CPU strain.
additionally simply timedValue to have RWMutex
to avoid concurrent calls to DiskInfo() getting
serialized, this has an effect on all calls that
use GetDiskInfo() on the same disks.
Such as getOnlineDisks, getOnlineDisksWithoutHealing
Current implementation requires server pools to have
same erasure stripe sizes, to facilitate same SLA
and expectations.
This PR allows server pools to be variadic, i.e they
do not have to be same erasure stripe sizes - instead
they should have SLA for parity ratio.
If the parity ratio cannot be guaranteed by the new
server pool, the deployment is rejected i.e server
pool expansion is not allowed.
additionally also configure http2 healthcheck
values to quickly detect unstable connections
and let them timeout.
also use single transport for proxying requests
Go stdlib resolver doesn't support caching DNS
resolutions, since we compile with CGO disabled
we are more probe to DNS flooding for all network
calls to resolve for DNS from the DNS server.
Under various containerized environments such as
VMWare this becomes a problem because there are
no DNS caches available and we may end up overloading
the kube-dns resolver under concurrent I/O.
To circumvent this issue implement a DNSCache resolver
which resolves DNS and caches them for around 10secs
with every 3sec invalidation attempted.
This PR fixes a hang which occurs quite commonly at higher concurrency
by allowing following changes
- allowing lower connections in time_wait allows faster socket open's
- lower idle connection timeout to ensure that we let kernel
reclaim the time_wait connections quickly
- increase somaxconn to 4096 instead of 2048 to allow larger tcp
syn backlogs.
fixes#10413
- select lockers which are non-local and online to have
affinity towards remote servers for lock contention
- optimize lock retry interval to avoid sending too many
messages during lock contention, reduces average CPU
usage as well
- if bucket is not set, when deleteObject fails make sure
setPutObjHeaders() honors lifecycle only if bucket name
is set.
- fix top locks to list out always the oldest lockers always,
avoid getting bogged down into map's unordered nature.
In `(*cacheObjects).GetObjectNInfo` copy the metadata before spawning a goroutine.
Clean up a few map[string]string copies as well, reducing allocs and simplifying the code.
Fixes#10426
MaxConnsPerHost can potentially hang a call without any
way to timeout, we do not need this setting for our proxy
and gateway implementations instead IdleConn settings are
good enough.
Also ensure to use NewRequestWithContext and make sure to
take the disks offline only for network errors.
Fixes#10304
with the merge of https://github.com/etcd-io/etcd/pull/11823
etcd v3.5.0 will now have a properly imported versioned path
this fixes our pending migration to newer repo
Use a separate client for these calls that can take a long time.
Add request context to these so they are canceled when the client
disconnects instead except for ListObject which doesn't have any equivalent.
- Implement a new xl.json 2.0.0 format to support,
this moves the entire marshaling logic to POSIX
layer, top layer always consumes a common FileInfo
construct which simplifies the metadata reads.
- Implement list object versions
- Migrate to siphash from crchash for new deployments
for object placements.
Fixes#2111
This value is requested on every upload when there are multiple zones.
Since this will result in an RPC call to every remote disk this scales
quite badly in a distributed setup. Load every 1second interval.
2 servers, localhost only. In large distributed setups much bigger
gains can be expected.
```
Operations: 21743 -> 22454
* Average: +3.28% (+0.0 MiB/s) throughput, +3.28% (+11.9) obj/s
* Fastest: +3.37% (+0.0 MiB/s) throughput, +3.37% (+13.0) obj/s
* 50% Median: +3.03% (+0.0 MiB/s) throughput, +3.03% (+11.2) obj/s
* Slowest: +8.03% (+0.0 MiB/s) throughput, +8.03% (+22.8) obj/s
```
For easy management of this a generic helper has been added.
global WORM mode is a complex piece for which
the time has passed, with the advent of S3 compatible
object locking and retention implementation global
WORM is sort of deprecated, this has been mentioned
in our documentation for some time, now the time
has come for this to go.
- keep long running obd network tests alive
- fix error - wrong number of parents in process OBD info
- ensure that osinfo does not error out when inside containers
- remove limit on max number of connections per client transport
The generic client transport uses a default limit of 64 conns per transport.
This could end up limiting and throttling usage, and artificially slowing
down the performance of MinIO even on hardware capable of doing better.
Too many deployments come up with an odd number
of hosts or drives, to facilitate even distribution
among those setups allow for odd and prime numbers
based packs.
- Implement a graph algorithm to test network bandwidth from every
node to every other node
- Saturate any network bandwidth adaptively, accounting for slow
and fast network capacity
- Implement parallel drive OBD tests
- Implement a paging mechanism for OBD test to provide periodic updates to client
- Implement Sys, Process, Host, Mem OBD Infos
- Add conservative timeouts upto 3 minutes
for internode communication
- Add aggressive timeouts of 30 seconds
for gateway communication
Fixes#9105Fixes#8732Fixes#8881Fixes#8376Fixes#9028
this PR enforces md5sum verification for following
API's to be compatible with AWS S3 spec
- PutObjectRetention
- PutObjectLegalHold
Co-authored-by: Harshavardhana <harsha@minio.io>
Allow downloading goroutine dump to help detect leaks
or overuse of goroutines.
Extensions are now type dependent.
Change `profiling` -> `profile` prefix, since that is what they are
not the abstract concept.
Add dummy calls which respond success when ACL's
are set to be private and fails, if user tries
to change them from their default 'private'
Some applications such as nuxeo may have an
unnecessary requirement for this operation,
we support this anyways such that don't have
to fully implement the functionality just that
we can respond with success for default ACLs
First step is to ensure that Path component is not decoded
by gorilla/mux to avoid routing issues while handling
certain characters while uploading through PutObject()
Delay the decoding and use PathUnescape() to escape
the `object` path component.
Thanks to @buengese and @ncw for neat test cases for us
to test with.
Fixes#8950Fixes#8647
For 'snapshot' type profiles, record a 'before' profile that can be used
as `go tool pprof -base=before ...` to compare before and after.
"Before" profiles are included in the zipped package.
[`runtime.MemProfileRate`](https://golang.org/pkg/runtime/#pkg-variables)
should not be updated while the application is running, so we set it at startup.
Co-authored-by: Harshavardhana <harsha@minio.io>
On every restart of the server, usage was being
calculated which is not useful instead wait for
sufficient time to start the crawling routine.
This PR also avoids lots of double allocations
through strings, optimizes usage of string builders
and also avoids crawling through symbolic links.
Fixes#8844
Enabling the memory profiling has a significant impact on performance.
Reduce the profiling rate by 2 orders of magnitude. It is still 128x smaller than default so it should be plenty.
object lock config is enabled for a bucket.
Creating a bucket with object lock configuration
enabled does not automatically cause WORM protection
to be applied. PUT operation needs to specifically
request object locking or bucket has to have default
retention settings configured.
Fixes regression introduced in #8657
Use reference format to initialize lockers
during startup, also handle `nil` for NetLocker
in dsync and remove *errorLocker* implementation
Add further tuning parameters such as
- DialTimeout is now 15 seconds from 30 seconds
- KeepAliveTimeout is not 20 seconds, 5 seconds
more than default 15 seconds
- ResponseHeaderTimeout to 10 seconds
- ExpectContinueTimeout is reduced to 3 seconds
- DualStack is enabled by default remove setting
it to `true`
- Reduce IdleConnTimeout to 30 seconds from
1 minute to avoid idleConn build up
Fixes#8773
In certain organizations policy claim names
can be not just 'policy' but also things like
'roles', the value of this field might also
be *string* or *[]string* support this as well
In this PR we are still not supporting multiple
policies per STS account which will require a
more comprehensive change.
This ensures that we can update the
- .minio.sys is updated for accounting/data usage purposes
- .minio.sys is updated to indicate if backend is encrypted
or not.
Admin data usage info API returns the following
(Only FS & XL, for now)
- Number of buckets
- Number of objects
- The total size of objects
- Objects histogram
- Bucket sizes
level - this PR builds on #8120 which
added PutBucketObjectLockConfiguration and
GetBucketObjectLockConfiguration APIS
This PR implements PutObjectRetention,
GetObjectRetention API and enhances
PUT and GET API operations to display
governance metadata if permissions allow.
This PR implements locking from a global entity into
a more localized set level entity, allowing for locks
to be held only on the resources which are writing
to a collection of disks rather than a global level.
In this process this PR also removes the top-level
limit of 32 nodes to an unlimited number of nodes. This
is a precursor change before bring in bucket expansion.
- adding oauth support to MinIO browser (#8400) by @kanagaraj
- supports multi-line get/set/del for all config fields
- add support for comments, allow toggle
- add extensive validation of config before saving
- support MinIO browser to support proper claims, using STS tokens
- env support for all config parameters, legacy envs are also
supported with all documentation now pointing to latest ENVs
- preserve accessKey/secretKey from FS mode setups
- add history support implements three APIs
- ClearHistory
- RestoreHistory
- ListHistory
- add help command support for each config parameters
- all the bug fixes after migration to KV, and other bug
fixes encountered during testing.
net.Error is very unreliable in providing better error
handling, we need to ensure that we always have a fallback
option in case of network failures.
This fixes an important issue in our distributed server
setups when one of the servers is down, all deployments
out there are recommended to upgrade after this fix is
merged to ensure that availability is not lost.
Fixes#8127Fixes#8016Fixes#7964
In situations such as when client uploading data,
prematurely disconnects from server such as pressing
ctrl-c before uploading all the data. Under this
situation in distributed setup we prematurely
disconnect disks causing a reconnect loop. This has
an adverse affect we end up leaving a lot of files
in temporary location which ideally should have been
cleaned up when Put() prematurely fails.
This is also a regression which got introduced in #7610
This allows for canonicalization of the strings
throughout our code and provides a common space
for all these constants to reside.
This list is rather non-exhaustive but captures
all the headers used in AWS S3 API operations
r.URL.Path is empty when HEAD bucket with virtual
DNS requests come in since bucket is now part of
r.Host, we should use our domain names and fetch
the right bucket/object names.
This fixes an really old issue in our federation
setups.
Bulk delete at storage level in Multiple Delete Objects API
In order to accelerate bulk delete in Multiple Delete objects API,
a new bulk delete is introduced in storage layer, which will accept
a list of objects to delete rather than only one. Consequently,
a new API is also need to be added to Object API.
Deprecate the use of Admin Peers concept and migrate all peer
communication to Notification subsystem. This finally allows
for a common subsystem for all peer notification in case of
distributed server deployments.
One user reported having discovered the following error:
API: SYSTEM()
Time: 20:06:17 UTC 12/06/2018
Error: xml: encoding "US-ASCII" declared but Decoder.CharsetReader is nil
1: cmd/handler-utils.go:43:cmd.parseLocationConstraint()
2: cmd/auth-handler.go:250:cmd.checkRequestAuthType()
3: cmd/bucket-handlers.go:411:cmd.objectAPIHandlers.PutBucketHandler()
4: cmd/api-router.go100cmd.(objectAPIHandlers).PutBucketHandler-fm()
5: net/http/server.go:1947:http.HandlerFunc.ServeHTTP()
Hence, adding support of different xml encoding. Although there
is no clear specification about it, even setting "GARBAGE" as an xml
encoding won't change the behavior of AWS, hence the encoding seems
to be ignored.
This commit will follow that behavior and will ignore encoding field
and consider all xml as utf8 encoded.
This refactor brings a change which allows
targets to be added in a cleaner way and also
audit is now moved out.
This PR also simplifies logger dependency for auditing
This PR brings an additional logger implementation
called AuditLog which logs to http targets
The intention is to use AuditLog to log all incoming
requests, this is used as a mechanism by external log
collection entities for processing Minio requests.
Two handlers are added to admin API to enable profiling and disable
profiling of a server in a standalone mode, or all nodes in the
distributed mode.
/minio/admin/profiling/start/{cpu,block,mem}:
- Start profiling and return starting JSON results, e.g. one
node is offline.
/minio/admin/profiling/download:
- Stop the on-going profiling task
- Stream a zip file which contains all profiling files that can
be later inspected by go tool pprof
Currently Go http connection pool was not being properly
utilized leading to degrading performance as the number
of concurrent requests increased.
As recommended by Go implementation, we have to drain the
response body and close it.
Currently, requestid field in logEntry is not populated, as the
requestid field gets set at the very end.
It is now set before regular handler functions. This is also
useful in setting it as part of the XML error response.
Travis build for ppc64le has been quite inconsistent and stays queued
for most of the time. Removing this build as part of Travis.yml for
the time being.
With CoreDNS now supporting etcdv3 as the DNS backend, we
can update our federation target to etcdv3. Users will now be
able to use etcdv3 server as the federation backbone.
Minio will update bucket data to etcdv3 and CoreDNS can pick
that data up and serve it as bucket style DNS path.
This PR adds disk based edge caching support for minio server.
Cache settings can be configured in config.json to take list of disk drives,
cache expiry in days and file patterns to exclude from cache or via environment
variables MINIO_CACHE_DRIVES, MINIO_CACHE_EXCLUDE and MINIO_CACHE_EXPIRY
Design assumes that Atime support is enabled and the list of cache drives is
fixed.
- Objects are cached on both GET and PUT/POST operations.
- Expiry is used as hint to evict older entries from cache, or if 80% of cache
capacity is filled.
- When object storage backend is down, GET, LIST and HEAD operations fetch
object seamlessly from cache.
Current Limitations
- Bucket policies are not cached, so anonymous operations are not supported in
offline mode.
- Objects are distributed using deterministic hashing among list of cache
drives specified.If one or more drives go offline, or cache drive
configuration is altered - performance could degrade to linear lookup.
Fixes#4026
This commit adds the bucket delete and bucket policy functionalities
to the browser.
Part of rewriting the browser code to follow best practices and
guidelines of React (issues #5409 and #5410)
The backend code has been modified by @krishnasrinivas to prevent
issue #4498 from occuring. The relevant changes have been made to the
code according to the latest commit and the unit tests in the backend.
This commit also addresses issue #5449.
- Data from disk was being read after bitrot verification to return
data for GetObject. Strictly speaking this does not guarantee bitrot
protection, as disks may return bad data even temporarily.
- This fix reads data from disk, verifies data for bitrot and then
returns data to the client directly.
MaxIdleConns limits the total number of connections
kept in the pool for re-use. In addition, MaxIdleConnsPerHost
limits the number for a single host. Since minio gateways
usually connect to the same host, setting `MaxIdleConns = 100`
won't really have much of an impact since the idle connection
pool is limited to 2 anyway.
Now, with the pool set to a limit of 2, and when using
the client heavily from 2+ goroutines, the `http.Transport`
will open a connection, use it, then try to return it to
the idle-pool which often fails since there's a limit of 2.
So it's going to close the connection and new ones will be
opened on demand again, many of which get closed soon after
being used. Since those connections/sockets don't disappear
from the OS immediately, use `MaxIdleConnsPerHost = 100`
which fixes this problem.
Save http trace to a file instead of displaying it onto the console.
the environment variable MINIO_HTTP_TRACE will be a filepath instead
of a boolean.
This to handle the scenario where both json and http tracing are
turned on. In that case, both http trace and json output are displayed
on the screen making the json not parsable. Loging this trace onto
a file helps us avoid that scenario.
Fixes#5263
Previously ID/ETag from backend service is used as is which causes
failure on s3cmd like tools where those tools use ETag as checksum to
validate data. This is fixed by prepending "-1".
Refer minio/mint#193minio/mint#201
The following commit f44f2e341c
fix was incomplete and we still had presigned URLs printing
in query strings in wrong fashion.
This PR fixes this properly. Avoid double encoding
percent encoded strings such as
`s3%!!(MISSING)A(MISSING)`
Print properly as json encoded.
`s3%3AObjectCreated%3A%2A`
This change adopts the upstream fix in this regard at
https://go-review.googlesource.com/#/c/41834/ for Minio's
purposes.
Go's current os.Stat() lacks support for lot of strange
windows files such as
- share symlinks on SMB2
- symlinks on docker nanoserver
- de-duplicated files on NTFS de-duplicated volume.
This PR attempts to incorporate the change mentioned here
https://blogs.msdn.microsoft.com/oldnewthing/20100212-00/?p=14963/
The article suggests to use Windows I/O manager to
dereference the symbolic link.
Fixes#4122
Ignore any network errors when registering a webhook
notifier during Minio startup sequence. This way server
can be started even if the webhook endpoint is not available
and unreachable.
url.Parse() wrongly parses an address of format "address:port"
which is fixed in go1.8. This inculcates a breaking change
on our end. We should fix this wrong usage everywhere so that
migrating to go1.8 eventually becomes smoother.
checkURL() is a generic function to check if a passed address
is valid. This commit adds support for addresses like `m1`
and `172.16.3.1` which is needed in MySQL and NATS. This commit
also adds tests.
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 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.
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
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.
- 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.
For command line arguments we are currently following
- <node-1>:/path ... <node-n>:/path
This patch changes this to
- http://<node-1>/path ... http://<node-n>/path
This makes sure that when SSL is enabled (for FS/single node mode),
the server address is picked up from the --address option (that needs
to include the hostname for SSL verification, and has to be input
appropriately by user), instead of just using ":<port>".
* Implements a Peer RPC router that sends info to all Minio servers in the cluster.
* Bucket notifications are propagated to all nodes via this RPC router.
* Bucket listener configuration is persisted to separate object layer
file (`listener.json`) and peer RPCs are used to communicate changes
throughout the cluster.
* When events are generated, RPC calls to send them to other servers
where bucket listeners may be connected is implemented.
* Some bucket notification tests are now disabled as they cannot work in
the new design.
* Minor fix in `funcFromPC` to use `path.Join`
* Test code for controller-handler operations:
* Heal operations
* List operation
* Switch to "testing" lib, moving away from gocheck
* Minor refactors
* Remove extra call to initGracefulShutdown
* Remove dead code in mainControl:
Dead code found by the TestControlMain() test function that always
passes.
* Add tests for control-*-main.go
