Replace the `io.Pipe` from streamingBitrotWriter -> CreateFile with a fixed size ring buffer.
This will add an output buffer for encoded shards to be written to disk - potentially via RPC.
This will remove blocking when `(*streamingBitrotWriter).Write` is called, and it writes hashes and data.
With current settings, the write looks like this:
```
Outbound
┌───────────────────┐ ┌────────────────┐ ┌───────────────┐ ┌────────────────┐
│ │ Parr. │ │ (http body) │ │ │ │
│ Bitrot Hash │ Write │ Pipe │ Read │ HTTP buffer │ Write (syscall) │ TCP Buffer │
│ Erasure Shard │ ──────────► │ (unbuffered) │ ────────────► │ (64K Max) │ ───────────────────► │ (4MB) │
│ │ │ │ │ (io.Copy) │ │ │
└───────────────────┘ └────────────────┘ └───────────────┘ └────────────────┘
```
We write a Hash (32 bytes). Since the pipe is unbuffered, it will block until the 32 bytes have
been delivered to the TCP buffer, and the next Read hits the Pipe.
Then we write the shard data. This will typically be bigger than 64KB, so it will block until two blocks
have been read from the pipe.
When we insert a ring buffer:
```
Outbound
┌───────────────────┐ ┌────────────────┐ ┌───────────────┐ ┌────────────────┐
│ │ │ │ (http body) │ │ │ │
│ Bitrot Hash │ Write │ Ring Buffer │ Read │ HTTP buffer │ Write (syscall) │ TCP Buffer │
│ Erasure Shard │ ──────────► │ (2MB) │ ────────────► │ (64K Max) │ ───────────────────► │ (4MB) │
│ │ │ │ │ (io.Copy) │ │ │
└───────────────────┘ └────────────────┘ └───────────────┘ └────────────────┘
```
The hash+shard will fit within the ring buffer, so writes will not block - but will complete after a
memcopy. Reads can fill the 64KB buffer if there is data for it.
If the network is congested, the ring buffer will become filled, and all syscalls will be on full buffers.
Only when the ring buffer is filled will erasure coding start blocking.
Since there is always "space" to write output data, we remove the parallel writing since we are
always writing to memory now, and the goroutine synchronization overhead probably not worth taking.
If the output were blocked in the existing, we would still wait for it to unblock in parallel write, so it would
make no difference there - except now the ring buffer smoothes out the load.
There are some micro-optimizations we could look at later. The biggest is that, in most cases,
we could encode directly to the ring buffer - if we are not at a boundary. Also, "force filling" the
Read requests (i.e., blocking until a full read can be completed) could be investigated and maybe
allow concurrent memory on read and write.
Metrics being added:
- read_tolerance: No of drive failures that can be tolerated without
disrupting read operations
- write_tolerance: No of drive failures that can be tolerated without
disrupting write operations
- read_health: Health of the erasure set in a pool for read operations
(1=healthy, 0=unhealthy)
- write_health: Health of the erasure set in a pool for write operations
(1=healthy, 0=unhealthy)
Adds regression test for #19699
Failures are a bit luck based, since it requires objects to be placed on different sets.
However this generates a failure prior to #19699
* Revert "Revert "Fix incorrect merging of slash-suffixed objects (#19699)""
This reverts commit f30417d9a8.
* Don't override when suffix doesn't match. Instead rely on quorum for each.
Instead of having "online" and "healing" as two metrics, replace with a
single metric "health" which can have following values:
0 = offline
1 = healthy
2 = healing
If two objects share everything but one object has a slash prefix, those would be merged in listings,
with secondary properties used for a tiebreak.
Example: An object with the key `prefix/obj` would be merged with an object named `prefix/obj/`.
While this violates the [no object can be a prefix of another](https://min.io/docs/minio/linux/operations/concepts/thresholds.html#conflicting-objects), let's resolve these.
If we have an object with 'name' and a directory named 'name/' discard the directory only - but allow objects
of 'name' and 'name/' (xldir) to be uniquely returned.
Regression from #15772
canceled callers might linger around longer,
can potentially overwhelm the system. Instead
provider a caller context and canceled callers
don't hold on to them.
Bonus: we have no reason to cache errors, we should
never cache errors otherwise we can potentially have
quorum errors creeping in unexpectedly. We should
let the cache when invalidating hit the actual resources
instead.
LastPong is saved as nanoseconds after a connection or reconnection but
saved as seconds when receiving a pong message. The code deciding if
a pong is too old can be skewed since it assumes LastPong is only in
seconds.
Accept multipart uploads where the combined checksum provides the expected part count.
It seems this was added by AWS to make the API more consistent, even if the
data is entirely superfluous on multiple levels.
Improves AWS S3 compatibility.
This commit adds support for MinKMS. Now, there are three KMS
implementations in `internal/kms`: Builtin, MinIO KES and MinIO KMS.
Adding another KMS integration required some cleanup. In particular:
- Various KMS APIs that haven't been and are not used have been
removed. A lot of the code was broken anyway.
- Metrics are now monitored by the `kms.KMS` itself. For basic
metrics this is simpler than collecting metrics for external
servers. In particular, each KES server returns its own metrics
and no cluster-level view.
- The builtin KMS now uses the same en/decryption implemented by
MinKMS and KES. It still supports decryption of the previous
ciphertext format. It's backwards compatible.
- Data encryption keys now include a master key version since MinKMS
supports multiple versions (~4 billion in total and 10000 concurrent)
per key name.
Signed-off-by: Andreas Auernhammer <github@aead.dev>
If used, 'opts.Marker` will cause many missed entries since results are returned
unsorted, and pools are serialized.
Switch to fully concurrent listing and merging across pools to return sorted entries.
It is expected that whoever is using the credentials which has
the proper set of permissions must be able to run.
`mc support perf object`
While the root login is disabled.
fixes#19648
AWS S3 returns the actual object size as part of XML
response for InvalidRange error, this is used apparently
by SDKs to retry the request without the range.
'opts.Marker` is causing many missed entries if used since results are returned unsorted. Also since pools are serialized.
Switch to do fully concurrent listing and merging across pools to return sorted entries.
Returning errors on listings is impossible with the current API, so document that.
Return an error at once if no drives are found instead of just returning an empty listing and no error.
This is to support deployments migrating from a multi-pooled
wider stripe to lower stripe. MINIO_STORAGE_CLASS_STANDARD
is still expected to be same for all pools. So you can satisfy
adding custom drive count based pools by adjusting the storage
class value.
```
version: v2
address: ':9000'
rootUser: 'minioadmin'
rootPassword: 'minioadmin'
console-address: ':9001'
pools: # Specify the nodes and drives with pools
-
args:
- 'node{11...14}.example.net/data{1...4}'
-
args:
- 'node{15...18}.example.net/data{1...4}'
-
args:
- 'node{19...22}.example.net/data{1...4}'
-
args:
- 'node{23...34}.example.net/data{1...10}'
set-drive-count: 6
```
ILM actions due to ExpiredObjectDeleteAllVersions and
DelMarkerExpiration are ignored when object locking is enabled on a
bucket.
Note: This applies to object versions which may not have retention
configured on them. This applies to all object versions in this bucket,
including those created before the retention config was applied.