Removes the bloom filter since it has so limited usability, often gets saturated anyway and adds a bunch of complexity to the scanner.
Also removes a tiny bit of CPU by each write operation.
xl.meta gets written and never rolled back, however
we definitely need to validate the state that is
persisted on the disk, if there are inconsistencies
- more than write quorum we should return an error
to the client
- if write quorum was achieved however there are
inconsistent xl.meta's we should simply trigger
an MRF on them
inlined data often is bigger than the allowed
O_DIRECT alignment, so potentially we can write
'xl.meta' without O_DSYNC instead we can rely on
O_DIRECT + fdatasync() instead.
This PR allows O_DIRECT on inlined data that
would gain the benefits of performing O_DIRECT,
eventually performing an fdatasync() at the end.
Performance boost can be observed here for small
objects < 128KiB. The performance boost is mainly
seen on HDD, and marginal on NVMe setups.
a/b/c/d/ where `a/b/c/` exists results in additional syscalls
such as an Lstat() call to verify if the `a/b/c/` exists
and its a directory.
We do not need to do this on MinIO since the parent prefixes
if exist, we can simply return success without spending
additional syscalls.
Also this implementation attempts to simply use Access() calls
to avoid os.Stat() calls since the latter does memory allocation
for things we do not need to use.
Access() is simpler since we have a predictable structure on
the backend and we know exactly how our path structures are.
Do completely independent multipart uploads.
In distributed mode, a lock was held to merge each multipart
upload as it was added. This lock was highly contested and
retries are expensive (timewise) in distributed mode.
Instead, each part adds its metadata information uniquely.
This eliminates the per object lock required for each to merge.
The metadata is read back and merged by "CompleteMultipartUpload"
without locks when constructing final object.
Co-authored-by: Harshavardhana <harsha@minio.io>
fix: change timedvalue to return previous cached value
caller can interpret the underlying error and decide
accordingly, places where we do not interpret the
errors upon timedValue.Get() - we should simply use
the previously cached value instead of returning "empty".
Bonus: remove some unused code
We need to make sure if we cannot read bucket metadata
for some reason, and bucket metadata is not missing and
returning corrupted information we should panic such
handlers to disallow I/O to protect the overall state
on the system.
In-case of such corruption we have a mechanism now
to force recreate the metadata on the bucket, using
`x-minio-force-create` header with `PUT /bucket` API
call.
Additionally fix the versioning config updated state
to be set properly for the site replication healing
to trigger correctly.
The test expects from DeleteFile to return errDiskNotFound when the disk
is not available. It calls os.RemoveAll() to remove one disk after XL storage
initialization. However, this latter contains some goroutines which can
race with os.RemoveAll() and then the test fails sporadically with
returning random errors.
The commit will tweak the initialization routine of the XL storage to
only run deletion of temporary and metacache data in the background,
so TestXLStorageDeleteFile won't fail anymore.
heal bucket metadata and IAM entries for
sites participating in site replication from
the site with the most updated entry.
Co-authored-by: Harshavardhana <harsha@minio.io>
Co-authored-by: Aditya Manthramurthy <aditya@minio.io>
This PR fixes two issues
- The first fix is a regression from #14555, the fix itself in #14555
is correct but the interpretation of that information by the
object layer code for "replication" was not correct. This PR
tries to fix this situation by making sure the "Delete" replication
works as expected when "VersionPurgeStatus" is already set.
Without this fix, there is a DELETE marker created incorrectly on
the source where the "DELETE" was triggered.
- The second fix is perhaps an older problem started since we inlined-data
on the disk for small objects, CopyObject() incorrectly inline's
a non-inlined data. This is due to the fact that we have code where
we read the `part.1` under certain conditions where the size of the
`part.1` is less than the specific "threshold".
This eventually causes problems when we are "deleting" the data that
is only inlined, which means dataDir is ignored leaving such
dataDir on the disk, that looks like an inconsistent content on
the namespace.
fixes#14767
```
tmp = buf[want:]
```
Would potentially crash when `buf` is truncated for some reason
and does not have the expected bytes, this is of course considered
not normal and is an odd situation. But we do not need to crash
here instead allow for errors to be returned and let callers handle
the errors.
This PR simply adds a warning message when it detects older kernel
versions and warn's them about potential performance issues on this
kernel.
The issue can be seen only with parallel I/O across all drives
on denser setups such as 90 drives or 45 drives per server configurations.
The main goal of this PR is to solve the situation where disks stop
responding to operations. This generally causes an FD build-up and
eventually will crash the server.
This adds detection of hung disks, where calls on disk get stuck.
We add functionality to `xlStorageDiskIDCheck` where it keeps
track of the number of concurrent requests on a given disk.
A total number of 100 operations are allowed. If this limit is reached
we will block (but not reject) new requests, but we will monitor the
state of the disk.
If no requests have been completed or updated within a 15-second
window, we mark the disk as offline. Requests that are blocked will be
unblocked and return an error as "faulty disk".
New requests will be rejected until the disk is marked OK again.
Once a disk has been marked faulty, a check will run every 5 seconds that
will attempt to write and read back a file. As long as this fails the disk will
remain faulty.
To prevent lots of long-running requests to mark the disk faulty we
implement a callback feature that allows updating the status as parts
of these operations are running.
We add a reader and writer wrapper that will update the status of each
successful read/write operation. This should allow fine enough granularity
that a slow, but still operational disk will not reach 15 seconds where
50 operations have not progressed.
Note that errors themselves are not enough to mark a disk faulty.
A nil (or io.EOF) error will mark a disk as "good".
* Make concurrent disk setting configurable via `_MINIO_DISK_MAX_CONCURRENT`.
* de-couple IsOnline() from disk health tracker
The purpose of IsOnline() is to ensure that we
reconnect the drive only when the "drive" was
- disconnected from network we need to validate
if the drive is "correct" and is the same drive
which belongs to this server.
- drive was replaced we have to format it - we
support hot swapping of the drives.
IsOnline() is not meant for taking the drive offline
when it is hung, it is not useful we can let the
drive be online instead "return" errors for relevant
calls.
* return errFaultyDisk for DiskInfo() call
Co-authored-by: Harshavardhana <harsha@minio.io>
Possible future Improvements:
* Unify the REST server and local xlStorageDiskIDCheck. This would also improve stats significantly.
* Allow reads/writes to be aborted by the context.
* Add usage stats, concurrent count, blocked operations, etc.
This is a side-affect of the optimization done in PR #13544 which
causes a certain type of delete operations on given object versions
can cause lastVersion indication to be skipped, which leads to
an `xl.meta` where Versions[] slice is empty while the entire
file is intact by itself.
This PR tries to ensure that such files are visible and deletable
by regular means of listing as null 'delete-marker' and also
avoid the situation where this potential issue might arise.
- speedtest logs calls that were canceled
spuriously, in situations where it should
be ignored.
- all errors of interest are always sent back
to the client there is no need to log them
on the server console.
- PUT failures should negate the increments
such that GET is not attempted on unsuccessful
calls.
- do not attempt MRF on speedtest objects.
In the testing mode, reformatting disks will fail because the healing
code will complain if one disk is in root mode. This commit will
automatically set all disks as non-root if MINIO_CI_CD is set.
The current code considers a pool with all root disks to be as part
of a testing environment even if there are other pools with mounted
disks. This will result to illegitimate writing in root disks.
Fix this by simplifing the logic: require MINIO_CI_CD in order to skip
root disk check.
startup speed-up, currently getFormatErasureInQuorum()
would spend up to 2-3secs when there are 3000+ drives
for example in a setup, simplify this implementation
to use drive counts.
This PR removes an unnecessary state that gets
passed around for DiskIDs, which is not necessary
since each disk exactly knows which pool and which
set it belongs to on a running system.
Currently cached DiskId's won't work properly
because it always ends up skipping offline disks
and never runs healing when disks are offline, as
it expects all the cached diskIDs to be present
always. This also sort of made things in-flexible
in terms perhaps a new diskID for `format.json`.
(however this is not a big issue)
This is an unnecessary requirement that healing
via scanner needs all drives to be online, instead
healing should trigger even when partial nodes
and drives are available this ensures that we
keep the SLA in-tact on the objects when disks
are offline for a prolonged period of time.
- create internal erasure volumes only if the disk is unformatted
- return a copy of format data in xlStorage.ReadAll
- parse env vars only once, to be re-used by xl-storage
This speed-up is intended for faster startup times
for almost all MinIO operations. Changes here are
- Drives are not re-read for 'format.json' on a regular
basis once read during init is remembered and refreshed
at 5 second intervals.
- Do not do O_DIRECT tests on drives with existing 'format.json'
only fresh setups need this check.
- Parallelize initializing erasureSets for multiple sets.
- Avoid re-reading format.json when migrating 'format.json'
from really old V1->V2->V3
- Keep a copy of local drives for any given server in memory
for a quick lookup.
repeated reads on single large objects in HPC like
workloads, need the following option to disable
O_DIRECT for a more effective usage of the kernel
page-cache.
However this optional should be used in very specific
situations only, and shouldn't be enabled on all
servers.
NVMe servers benefit always from keeping O_DIRECT on.
When reading input for PutObject or PutObjectPart add a readahead buffer for big inputs.
This will make network reads+hashing separate run async with erasure coding and writes. This will reduce overall latency in distributed setups where the input is from upstream and writes go to other servers.
We will read at 2 buffers ahead, meaning one will always be ready/waiting and one is currently being read from.
This improves PutObject and PutObjectParts for these cases.
When deleting multiple versions it "gives" up with an errFileVersionNotFound if
a version cannot be found. This effectively skips deleting other versions
sent in the same request.
This can happen on inconsistent objects. We should ignore errFileVersionNotFound
and continue with others.
We already ignore these at the caller level, this PR is continuation of 54a9877
This PR simplifies few things
- Multipart parts are renamed, upon failure are unrenamed() keep this
multipart specific behavior it is needed and works fine.
- AbortMultipart should blindly delete once lock is acquired instead
of re-reading metadata and calculating quorum, abort is a delete()
operation and client has no business looking for errors on this.
- Skip Access() calls to folders that are operating on
`.minio.sys/multipart` folder as well.
Large clusters with multiple sets, or multi-pool setups at times might
fail and report unexpected "file not found" errors. This can become
a problem during startup sequence when some files need to be created
at multiple locations.
- This PR ensures that we nil the erasure writers such that they
are skipped in RenameData() call.
- RenameData() doesn't need to "Access()" calls for `.minio.sys`
folders they always exist.
- Make sure PutObject() never returns ObjectNotFound{} for any
errors, make sure it always returns "WriteQuorum" when renameData()
fails with ObjectNotFound{}. Return appropriate errors for all
other cases.
data shards were wrong due to a healing bug
reported in #13803 mainly with unaligned object
sizes.
This PR is an attempt to automatically avoid
these shards, with available information about
the `xl.meta` and actually disk mtime.
single object delete was not working properly
on a bucket when versioning was suspended,
current version 'null' object was never removed.
added unit tests to cover the behavior
fixes#13783
dataDir loosely based on maxima is incorrect and does not
work in all situations such as disks in the following order
- xl.json migration to xl.meta there may be partial xl.json's
leftover if some disks are not yet connected when the disk
is yet to come up, since xl.json mtime and xl.meta is
same the dataDir maxima doesn't work properly leading to
quorum issues.
- its also possible that XLV1 might be true among the disks
available, make sure to keep FileInfo based on common quorum
and skip unexpected disks with the older data format.
Also, this PR tests upgrade from older to a newer release if the
data is readable and matches the checksum.
NOTE: this is just initial work we can build on top of this to do further tests.
This unit allows users to limit the maximum number of noncurrent
versions of an object.
To enable this rule you need the following *ilm.json*
```
cat >> ilm.json <<EOF
{
"Rules": [
{
"ID": "test-max-noncurrent",
"Status": "Enabled",
"Filter": {
"Prefix": "user-uploads/"
},
"NoncurrentVersionExpiration": {
"MaxNoncurrentVersions": 5
}
}
]
}
EOF
mc ilm import myminio/mybucket < ilm.json
```
Existing:
```go
type xlMetaV2 struct {
Versions []xlMetaV2Version `json:"Versions" msg:"Versions"`
}
```
Serialized as regular MessagePack.
```go
//msgp:tuple xlMetaV2VersionHeader
type xlMetaV2VersionHeader struct {
VersionID [16]byte
ModTime int64
Type VersionType
Flags xlFlags
}
```
Serialize as streaming MessagePack, format:
```
int(headerVersion)
int(xlmetaVersion)
int(nVersions)
for each version {
binary blob, xlMetaV2VersionHeader, serialized
binary blob, xlMetaV2Version, serialized.
}
```
xlMetaV2VersionHeader is <= 30 bytes serialized. Deserialized struct
can easily be reused and does not contain pointers, so efficient as a
slice (single allocation)
This allows quickly parsing everything as slices of bytes (no copy).
Versions are always *saved* sorted by modTime, newest *first*.
No more need to sort on load.
* Allows checking if a version exists.
* Allows reading single version without unmarshal all.
* Allows reading latest version of type without unmarshal all.
* Allows reading latest version without unmarshal of all.
* Allows checking if the latest is deleteMarker by reading first entry.
* Allows adding/updating/deleting a version with only header deserialization.
* Reduces allocations on conversion to FileInfo(s).
legacy objects in 'xl.json' after upgrade, should have
following sequence of events - bucket should have versioning
enabled and the object should have been overwritten with
another version of an object.
this situation was not handled, which would lead to older
objects to stay perpetually with "legacy" dataDir, however
these objects were readable by all means - there weren't
converted to newer format.
This PR fixes this situation properly.
- remove some duplicated code
- reported a bug, separately fixed in #13664
- using strings.ReplaceAll() when needed
- using filepath.ToSlash() use when needed
- remove all non-Go style comments from the codebase
Co-authored-by: Aditya Manthramurthy <donatello@users.noreply.github.com>
deleting collection of versions belonging
to same object, we can avoid re-reading
the xl.meta from the disk instead purge
all the requested versions in-memory,
the tradeoff is to allocate a map to de-dup
the versions, allow disks to be read only
once per object.
additionally reduce the data transfer between
nodes by shortening msgp data values.
offset+length should match the Size() of the individual parts
return 'errFileCorrupt' otherwise, to trigger healing of the individual
parts do not error out prematurely when healing such bitrot's upon
successful parts being written to the client.
another issue this PR fixes is to not return and error to
the client if we have just triggered a heal on a specific
part of the object, instead continue to read all the content
and let the heal happen asynchronously later.
This PR brings two optimizations mainly
for page-cache build-up and how to avoid
getting OOM killed in the process. Although
these memories are reclaimable Linux is not
fast enough to reclaim them as needed on a
very busy system. fadvise is a system call
implemented in Linux to advise page-cache to
avoid overload as we get significant amount
of requests on the server.
- FADV_SEQUENTIAL tells that all I/O from now
is going to be sequential, allowing for more
resposive throughput.
- FADV_NOREUSE tells kernel to start removing
things for this 'fd' from page-cache.
DeleteObject() on existing objects before `xl.json` to
`xl.meta` change were not working, not sure when this
regression was added. This PR fixes this properly.
Also this PR ensures that we perform rename of xl.json
to xl.meta only during "write" phase of the call i.e
either during Healing or PutObject() overwrites.
Also handles few other scenarios during migration where
`backendEncryptedFile` was missing deleteConfig() will
fail with `configNotFound` this case was not ignored,
which can lead to failure during upgrades.
once we have competed for locks, verify if the
context is still valid - this is to ensure that
we do not start readdir() or read() calls on the
drives on canceled connections.
This commit brings two locks instead of single lock for
WalkDir() calls on top of c25816eabc.
The main reason is to avoid contention between readMetadata()
and ListDir() calls, ListDir() can take time on prefixes that
are huge for readdir() but this shouldn't end up blocking
all readMetadata() operations, this allows for more room for
I/O while not overly penalizing all listing operations.
When unable to load existing metadata new versions
would not be written. This would leave objects in a
permanently unrecoverable state
Instead, start with clean metadata and write the incoming data.
Use a single allocation for reading the file, not the growing buffer of `io.ReadAll`.
Reuse the write buffer if we can when writing metadata in RenameData.
- remove sourceCh usage from healing
we already have tasks and resp channel
- use read locks to lookup globalHealConfig
- fix healing resolver to pick candidates quickly
that need healing, without this resolver was
unexpectedly skipping.
Synchronize bucket cycles so it is much more
likely that the same prefixes will be picked up
for scanning.
Use the global bloom filter cycle for that.
Bump bloom filter versions to clear those.
Objects uploaded in this format for example
```
mc cp /etc/hosts alias/bucket/foo/bar/xl.meta
mc ls -r alias/bucket/foo/bar
```
Won't list the object, handle this scenario.
We are observing heavy system loads, potentially
locking the system up for periods when concurrent
listing operations are performed.
We place a per-disk lock on walk IO operations.
This will minimize the impact of concurrent listing
operations on the entire system and de-prioritize
them compared to other operations.
Single list operations should remain largely unaffected.
Use `readMetadata` when reading version
information without data requested.
Reduces IO on inlined data.
Bonus: Inline compressed data as well when
compression is enabled.
- avoid extra lookup for 'xl.meta' since we are
definitely sure that it doesn't exist.
- use this in newMultipartUpload() as well
- also additionally do not write with O_DSYNC
to avoid loading the drives, instead create
'xl.meta' for listing operations without
O_DSYNC since these are ephemeral objects.
- do the same with newMultipartUpload() since
it gets synced when the PutObjectPart() is
attempted, we do not need to tax newMultipartUpload()
instead.
we will allow situations such as
```
a/b/1.txt
a/b
```
and
```
a/b
a/b/1.txt
```
we are going to document that this usecase is
not supported and we will never support it, if
any application does this users have to delete
the top level parent to make sure namespace is
accessible at lower level.
rest of the situations where the prefixes get
created across sets are supported as is.
destination path and old path will be similar
when healing occurs, this can lead to healed
parts being again purged leading to always an
inconsistent state on an object which might
further cause reduction in quorum eventually.
- remove use of getOnlineDisks() instead rely on fallbackDisks()
when disk return errors like diskNotFound, unformattedDisk
use other fallback disks to list from, instead of paying the
price for checking getOnlineDisks()
- optimize getDiskID() further to avoid large write locks when
looking formatLastCheck time window
This new change allows for a more relaxed fallback for listing
allowing for more tolerance and also eventually gain more
consistency in results even if using '3' disks by default.
Download files from *any* bucket/path as an encrypted zip file.
The key is included in the response but can be separated so zip
and the key doesn't have to be sent on the same channel.
Requires https://github.com/minio/pkg/pull/6
- it is possible that during I/O failures we might
leave partially written directories, make sure
we purge them after.
- rename current data-dir (null) versionId only after
the newer xl.meta has been written fully.
- attempt removal once for minioMetaTmpBucket/uuid/
as this folder is empty if all previous operations
were successful, this allows avoiding recursive os.Remove()
It makes sense that a node that has multiple disks starts when one
disk fails, returning an i/o error for example. This commit will make this
faulty tolerance available in this specific use case.
Also adding an API to allow resyncing replication when
existing object replication is enabled and the remote target
is entirely lost. With the `mc replicate reset` command, the
objects that are eligible for replication as per the replication
config will be resynced to target if existing object replication
is enabled on the rule.
This is to ensure that there are no projects
that try to import `minio/minio/pkg` into
their own repo. Any such common packages should
go to `https://github.com/minio/pkg`
This PR fixes two bugs
- Remove fi.Data upon overwrite of objects from inlined-data to non-inlined-data
- Workaround for an existing bug on disk with latest releases to ignore fi.Data
and instead read from the disk for non-inlined-data
- Addtionally add a reserved metadata header to indicate data is inlined for
a given version.
Previous PR #12351 added functions to read from the reader
stream to reduce memory usage, use the same technique in
few other places where we are not interested in reading the
data part.
multi-disk clusters initialize buffer pools
per disk, this is perhaps expensive and perhaps
not useful, for a running server instance. As this
may disallow re-use of buffers across sets,
this change ensures that buffers across sets
can be re-used at drive level, this can reduce
quite a lot of memory on large drive setups.
gracefully start the server, if there are other drives
available - print enough information for administrator
to notice the errors in console.
Bonus: for really large streams use larger buffer for
writes.
A cache structure will be kept with a tree of usages.
The cache is a tree structure where each keeps track
of its children.
An uncompacted branch contains a count of the files
only directly at the branch level, and contains link to
children branches or leaves.
The leaves are "compacted" based on a number of properties.
A compacted leaf contains the totals of all files beneath it.
A leaf is only scanned once every dataUsageUpdateDirCycles,
rarer if the bloom filter for the path is clean and no lifecycles
are applied. Skipped leaves have their totals transferred from
the previous cycle.
A clean leaf will be included once every healFolderIncludeProb
for partial heal scans. When selected there is a one in
healObjectSelectProb that any object will be chosen for heal scan.
Compaction happens when either:
- The folder (and subfolders) contains less than dataScannerCompactLeastObject objects.
- The folder itself contains more than dataScannerCompactAtFolders folders.
- The folder only contains objects and no subfolders.
- A bucket root will never be compacted.
Furthermore, if a has more than dataScannerCompactAtChildren recursive
children (uncompacted folders) the tree will be recursively scanned and the
branches with the least number of objects will be compacted until the limit
is reached.
This ensures that any branch will never contain an unreasonable amount
of other branches, and also that small branches with few objects don't
take up unreasonable amounts of space.
Whenever a branch is scanned, it is assumed that it will be un-compacted
before it hits any of the above limits. This will make the branch rebalance
itself when scanned if the distribution of objects has changed.
TLDR; With current values: No bucket will ever have more than 10000
child nodes recursively. No single folder will have more than 2500 child
nodes by itself. All subfolders are compacted if they have less than 500
objects in them recursively.
We accumulate the (non-deletemarker) version count for paths as well,
since we are changing the structure anyway.
MRF does not detect when a node is disconnected and reconnected quickly
this change will ensure that MRF is alerted by comparing the last disk
reconnection timestamp with the last MRF check time.
Signed-off-by: Anis Elleuch <anis@min.io>
Co-authored-by: Klaus Post <klauspost@gmail.com>
as there is no automatic way to detect if there
is a root disk mounted on / or /var for the container
environments due to how the root disk information
is masked inside overlay root inside container.
this PR brings an environment variable to set
root disk size threshold manually to detect the
root disks in such situations.
Part ETags are not available after multipart finalizes, removing this
check as not useful.
Signed-off-by: Poorna Krishnamoorthy <poorna@minio.io>
Co-authored-by: Harshavardhana <harsha@minio.io>
With this change, MinIO's ILM supports transitioning objects to a remote tier.
This change includes support for Azure Blob Storage, AWS S3 compatible object
storage incl. MinIO and Google Cloud Storage as remote tier storage backends.
Some new additions include:
- Admin APIs remote tier configuration management
- Simple journal to track remote objects to be 'collected'
This is used by object API handlers which 'mutate' object versions by
overwriting/replacing content (Put/CopyObject) or removing the version
itself (e.g DeleteObjectVersion).
- Rework of previous ILM transition to fit the new model
In the new model, a storage class (a.k.a remote tier) is defined by the
'remote' object storage type (one of s3, azure, GCS), bucket name and a
prefix.
* Fixed bugs, review comments, and more unit-tests
- Leverage inline small object feature
- Migrate legacy objects to the latest object format before transitioning
- Fix restore to particular version if specified
- Extend SharedDataDirCount to handle transitioned and restored objects
- Restore-object should accept version-id for version-suspended bucket (#12091)
- Check if remote tier creds have sufficient permissions
- Bonus minor fixes to existing error messages
Co-authored-by: Poorna Krishnamoorthy <poorna@minio.io>
Co-authored-by: Krishna Srinivas <krishna@minio.io>
Signed-off-by: Harshavardhana <harsha@minio.io>
