The blobstore implementation is entirely related to storagenode, so the
rightful place is together with the storagenode implementation.
Fixes https://github.com/storj/storj/issues/5754
Change-Id: Ie6637b0262cf37af6c3e558556c7604d9dc3613d
TestLoopContinuesAfterObserverError was failing due to system
granularity measuring the duration as 0.
TestDialer_DialTimeout was failing due to connection failure came with a
delay and wasn't being handled.
Change-Id: I4638c86f5d021a86c3d3529fab13cf3608f35c40
The threshold of piece deletions from the nodes during CommitObject
when overriding an existing object seemed to cause a race condition in
tests.
This change makes the threshold configurable so we can set it to maximum
so CommitObject waits until all pieces are removed from the nodes in the
test.
Change-Id: Idf6b52e71d0082a1cd87ad99a2edded6892d02a8
Piece deletion service was using KnownReliable method from
overlaycache to get nodes addresses to send delete request.
KnownReliable was always hitting DB because this method was
not using cache. This change is using new DownloadSelectionCache
to avoid direct DB calls.
Change is not perfect because DownloadSelectionCache is not as
precise as KnownReliable method and can select few more nodes
to which we will send delete request but difference should be
small and we can improve it later.
Updates https://github.com/storj/storj/issues/4959
Change-Id: I4c3d91089a18ac35ebcb469a56536c33f76e44ea
Currently slower storagenodes can slow down deletion queue.
To make piece deletion faster reduce the maximum time spent in
either dialing or piece deletion requests.
With this change:
* dial timeout is 3s
* request timeout is 15s
* fail threshold is set to 10min
Similarly, we'll mark storage node as failed when the timeout occurs.
The timeout usually indicates that the storagenode is overwhelmed.
Garbage collection will ensure that the pieces get deleted eventually.
Change-Id: Iec5de699f5917905f5807140e2c3252088c6399b
We are not using the benchmark results for anything, they are mostly
there to ensure that we don't break the benchmarks. So we can disable
CockroachDB for them.
Similarly add short versions of other tests.
Also try to precompile test/benchmark code.
Change-Id: I60b501789f70c289af68c37a052778dc75ae2b69
It's safer to create new connection pool for piece deletion
only if dialer have no existing pool assigned.
Change-Id: I26661683ab7c0198587905478057c01c8f533a7e
We want to enable connection pool for piece deletion to avoid
doing multiple SSL hanshakes to SN while massive deletion process.
Change-Id: Ic917e4eda304ee16a286926ef046fe9e38bf38ca
Satellites set their configuration values to default values using
cfgstruct, however, it turns out our tests don't test these values
at all! Instead, they have a completely separate definition system
that is easy to forget about.
As is to be expected, these values have drifted, and it appears
in a few cases test planet is testing unreasonable values that we
won't see in production, or perhaps worse, features enabled in
production were missed and weren't enabled in testplanet.
This change makes it so all values are configured the same,
systematic way, so it's easy to see when test values are different
than dev values or release values, and it's less hard to forget
to enable features in testplanet.
In terms of reviewing, this change should be actually fairly
easy to review, considering private/testplanet/satellite.go keeps
the current config system and the new one and confirms that they
result in identical configurations, so you can be certain that
nothing was missed and the config is all correct.
You can also check the config lock to see what actual config
values changed.
Change-Id: I6715d0794887f577e21742afcf56fd2b9d12170e
Make metainfo.RSConfig a valid pflag config value. This allows us to
configure the RSConfig as a string like k/m/o/n-shareSize, which makes
having multiple supported RS schemes easier in the future.
RS-related config values that are no longer needed have been removed
(MinTotalThreshold, MaxTotalThreshold, MaxBufferMem, Verify).
Change-Id: I0178ae467dcf4375c504e7202f31443d627c15e1
This PR fixes a deadlock that can happen when the number of piece
deletion requests is different from the distinct node count from those
requests. The success threshold should be based on the number of nodes
instead of the amount of requests
Change-Id: I83073a22eb1e111be1e27641cebcefecdc16afcb
services
This PR adds a limiter on the amount of concurrent objects deletion can be handled so
we don't run out of memory.
Change-Id: Id2ce368af6f86845fcdfd34cb2f5e460efe9b272
This change will require less work for the user of peiecedeletion
service by moving overlay database call into the package.
Change-Id: I14a150ab71fe885780e7a7a74db006a779507ae5
This runs each benchmark for one iteration to ensure that they are
valid. Unfortunately, it does not give any useful metrics as output.
Change-Id: I68940398c8dd849aed656bd12656f48d5df10128
Before the deleter would close its done channel once, so if additional
tests shared a storagenode, even if not in parallel, the later waits
would not work properly. This fixes that problem.
Change-Id: I7dcacf6699cef7c2c2948ba0f4369ef520601bf5
When running testplanet tests, mark storagenode peer PieceDeleter as in
testing mode so that you don't have to do it on each test.
Change-Id: I2592e02c63f8bcc9152ecf436bac4e798b08bccf
There was a race in the test code for piece deleter, which made it
possible to broadcast on the condition variable before anyone was
waiting. This change fixes that and has Wait take a context so it times
out with the context.
Change-Id: Ia4f77a7b7d2287d5ab1d7ba541caeb1ba036dba3
To improve delete performance, we want to process deletes asynchronously
once the message has been received from the satellite. This change makes
it so that storagenodes will send the delete request to a piece Deleter,
which will process a "best-effort" delete asynchronously and return a
success message to the satellite.
There is a configurable number of max delete workers and a max delete
queue size.
Change-Id: I016b68031f9065a9b09224f161b6783e18cf21e5
Currently it was possible that PopAll returns 1010 items, then
makes one RPC call with 1000 items, then RPC call 10 items. Meanwhile,
there have been added 500 new items added to the queue.
This change ensures that we pull items from the queue early and
try to make rpc batches as large as possible.
Change-Id: I1a30dde9164c2ff7b90c906a9544593c4f1cf0e9
During testing it's possible to get into a scenario where all nodes are
offline and list of requests is empty.
Change-Id: I271c0ca2c72009244df13e8bc1441fcd5f3da9e0
Instead of providing the database from outside to testplanet create it
inside and then allow wrapping and modifying it. This is more convenient
to use.
Change-Id: I9b8f69e6e0a19ff984b4e2bfe927c9100c77bc6c
This adds a piece deletion handler that has debounce for failed dialing
and batching multiple jobs into a single request.
Change-Id: If64021bebb2faae7f3e6bdcceef705aed41e7d7b
To handle concurrent deletion requests we need to combine them into a
single request.
To implement this we introduces few concurrency ideas:
* Combiner, which takes a node id and a Job and handles combining
multiple requests to a single batch.
* Job, which represents deleting of multiple piece ids with a
notification mechanism to the caller.
* Queue, which provides communication from Combiner to Handler.
It can limit the number of requests per work queue.
* Handler, which takes an active Queue and processes it until it has
consumed all the jobs.
It can provide limits to handling concurrency.
Change-Id: I3299325534abad4bae66969ffa16c6ed95d5574f
