This PR refactors the way we use buffers for O_DIRECT and
to re-use those buffers for messagepack reader writer.
After some extensive benchmarking found that not all objects
have this benefit, and only objects smaller than 64KiB see
this benefit overall.
Benefits are seen from almost all objects from
1KiB - 32KiB
Beyond this no objects see benefit with bulk call approach
as the latency of bytes sent over the wire v/s streaming
content directly from disk negate each other with no
remarkable benefits.
All other optimizations include reuse of msgp.Reader,
msgp.Writer using sync.Pool's for all internode calls.
The only purpose of check-dir flag in
ReadVersion is to return 404 when
an object has xl.meta but without data.
This is causing an extract call to the disk
which can be penalizing in case of busy system
where disks receive many concurrent access.
Add context to all (non-trivial) calls to the storage layer.
Contexts are propagated through the REST client.
- `context.TODO()` is left in place for the places where it needs to be added to the caller.
- `endWalkCh` could probably be removed from the walkers, but no changes so far.
The "dangerous" part is that now a caller disconnecting *will* propagate down, so a
"delete" operation will now be interrupted. In some cases we might want to disconnect
this functionality so the operation completes if it has started, leaving the system in a cleaner state.
- 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
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.
This PR adds jsoniter package to replace encoding/json
in places where faster json unmarshal is necessary
whenever input JSON is large enough.
Some benchmarking comparison between jsoniter and enconding/json
benchmark old MB/s new MB/s speedup
BenchmarkParseUnmarshal/N10-4 110.02 331.17 3.01x
BenchmarkParseUnmarshal/N100-4 125.74 524.09 4.17x
BenchmarkParseUnmarshal/N500-4 131.68 542.60 4.12x
BenchmarkParseUnmarshal/N1000-4 133.93 514.88 3.84x
BenchmarkParseUnmarshal/N5000-4 122.10 415.36 3.40x
BenchmarkParseUnmarshal/N10000-4 132.13 403.90 3.06x
This PR refactors object layer handling such
that upon failure in sub-system initialization
server reaches a stage of safe-mode operation
wherein only certain API operations are enabled
and available.
This allows for fixing many scenarios such as
- incorrect configuration in vault, etcd,
notification targets
- missing files, incomplete config migrations
unable to read encrypted content etc
- any other issues related to notification,
policies, lifecycle etc
- Heal if the part.1 is truncated from its original size
- Heal if the part.1 fails while being verified in between
- Heal if the part.1 fails while being at a certain offset
Other cleanups include make sure to flush the HTTP responses
properly from storage-rest-server, avoid using 'defer' to
improve call latency. 'defer' incurs latency avoid them
in our hot-paths such as storage-rest handlers.
Fixes#8319
It looks like from implementation point of view fastjson
parser pool doesn't behave the same way as expected
when dealing many `xl.json` from multiple disks.
The fastjson parser pool usage ends up returning incorrect
xl.json entries for checksums, with references pointing
to older entries. This led to the subtle bug where checksum
info is duplicated from a previous xl.json read of a different
file from different disk.
This is to avoid using unsafe.Pointer type
code dependency for MinIO, this causes
crashes on ARM64 platforms
Refer #8005 collection of runtime crashes due
to unsafe.Pointer usage incorrectly. We have
seen issues like this before when using
jsoniter library in the past.
This PR hopes to fix this using fastjson
Other listing optimizations include
- remove double sorting while filtering object entries
- improve error message when upload-id is not in quorum
- use jsoniter for full unmarshal json, instead of gjson
- remove unused code
This PR adds pass-through, single encryption at gateway and double
encryption support (gateway encryption with pass through of SSE
headers to backend).
If KMS is set up (either with Vault as KMS or using
MINIO_SSE_MASTER_KEY),gateway will automatically perform
single encryption. If MINIO_GATEWAY_SSE is set up in addition to
Vault KMS, double encryption is performed.When neither KMS nor
MINIO_GATEWAY_SSE is set, do a pass through to backend.
When double encryption is specified, MINIO_GATEWAY_SSE can be set to
"C" for SSE-C encryption at gateway and backend, "S3" for SSE-S3
encryption at gateway/backend or both to support more than one option.
Fixes#6323, #6696
On a heavily loaded server, getBucketInfo() becomes slow,
one can easily observe deleting an object causes many
additional network calls.
This PR is to let the underlying call return the actual
error and write it back to the client.
*) Add Put/Get support of multipart in encryption
*) Add GET Range support for encryption
*) Add CopyPart encrypted support
*) Support decrypting of large single PUT object
This PR implements an object layer which
combines input erasure sets of XL layers
into a unified namespace.
This object layer extends the existing
erasure coded implementation, it is assumed
in this design that providing > 16 disks is
a static configuration as well i.e if you started
the setup with 32 disks with 4 sets 8 disks per
pack then you would need to provide 4 sets always.
Some design details and restrictions:
- Objects are distributed using consistent ordering
to a unique erasure coded layer.
- Each pack has its own dsync so locks are synchronized
properly at pack (erasure layer).
- Each pack still has a maximum of 16 disks
requirement, you can start with multiple
such sets statically.
- Static sets set of disks and cannot be
changed, there is no elastic expansion allowed.
- Static sets set of disks and cannot be
changed, there is no elastic removal allowed.
- ListObjects() across sets can be noticeably
slower since List happens on all servers,
and is merged at this sets layer.
Fixes#5465Fixes#5464Fixes#5461Fixes#5460Fixes#5459Fixes#5458Fixes#5460Fixes#5488Fixes#5489Fixes#5497Fixes#5496
- Changes related to moving admin APIs
- admin APIs now have an endpoint under /minio/admin
- admin APIs are now versioned - a new API to server the version is
added at "GET /minio/admin/version" and all API operations have the
path prefix /minio/admin/v1/<operation>
- new service stop API added
- credentials change API is moved to /minio/admin/v1/config/credential
- credentials change API and configuration get/set API now require TLS
so that credentials are protected
- all API requests now receive JSON
- heal APIs are disabled as they will be changed substantially
- Heal API changes
Heal API is now provided at a single endpoint with the ability for a
client to start a heal sequence on all the data in the server, a
single bucket, or under a prefix within a bucket.
When a heal sequence is started, the server returns a unique token
that needs to be used for subsequent 'status' requests to fetch heal
results.
On each status request from the client, the server returns heal result
records that it has accumulated since the previous status request. The
server accumulates upto 1000 records and pauses healing further
objects until the client requests for status. If the client does not
request any further records for a long time, the server aborts the
heal sequence automatically.
A heal result record is returned for each entity healed on the server,
such as system metadata, object metadata, buckets and objects, and has
information about the before and after states on each disk.
A client may request to force restart a heal sequence - this causes
the running heal sequence to be aborted at the next safe spot and
starts a new heal sequence.
The problem in existing code was the following line
```
start := int(keyCrc%uint32(cardinality)) | 1
```
A given a value of N cardinality the ending result
because of the the bitwise '|' would lead to always
higher affinity to odd sequences.
As can be seen from the test cases that this can
lead to many objects being allocated the same set
of disks or atleast the first disk is an odd disk
always. This introduces a performance problem
for majority of the objects under concurrent load.
Remove `| 1` to provide a more cleaner distribution
and the new code will be.
```
start := int(keyCrc % uint32(cardinality))
```
Thanks to Krishna Srinivas for pointing out the bitwise
situation here.
Amazon S3 API expects all incoming stream has a content-length
set it was superflous for us to support object layer which supports
unknown sized stream as well, this PR removes such requirements
and explicitly error out if input stream is less than zero.
We don't need to typecast identifiers from
their base to type to same type again. This
is not a bug and compiler is fine to skip
it but it is better to avoid if not needed.
This change provides new implementations of the XL backend operations:
- create file
- read file
- heal file
Further this change adds table based tests for all three operations.
This affects also the bitrot algorithm integration. Algorithms are now
integrated in an idiomatic way (like crypto.Hash).
Fixes#4696Fixes#4649Fixes#4359
xl.storageDisks is sometimes passed to some low-level XL functions. Some disks in
xl.storageDisks are set to nil when they encounter some errors. This means all
elements in xl.storageDisks will be nil after some time which lead to an unusable XL.
This is an enhancement to the XL/distributed-XL mode. FS mode is
unaffected.
The ReadFileWithVerify storage-layer call is similar to ReadFile with
the additional functionality of performing bit-rot checking. It
accepts additional parameters for a hashing algorithm to use and the
expected hex-encoded hash string.
This patch provides significant performance improvement because:
1. combines the step of reading the file (during
erasure-decoding/reconstruction) with bit-rot verification;
2. limits the number of file-reads; and
3. avoids transferring the file over the network for bit-rot
verification.
ReadFile API is implemented as ReadFileWithVerify with empty hashing
arguments.
Credits to AB and Harsha for the algorithmic improvement.
Fixes#4236.
This PR also does backend format change to 1.0.1
from 1.0.0. Backward compatible changes are still
kept to read the 'md5Sum' key. But all new objects
will be stored with the same details under 'etag'.
Fixes#4312
Harshavardhana