We will be needing an infrequent chore to check which nodes are in the
reverify queue and synchronize that set with the 'contained' field in
the nodes db, since it is easily possible for them to get out of sync.
(We can't require that the reverification queue table be in the same
database as the nodes table, so maintaining consistency with SQL
transactions is out. Plus, even if they were in the same database, using
such SQL transactions to maintain consistency would be slow and
unwieldy.)
This commit adds the actual chore.
Refs: https://github.com/storj/storj/issues/5431
Change-Id: Id78b40bf69fae1ac39010e3b553315db8a1472bd
We will be needing an infrequent chore to check which nodes are in the
reverify queue and synchronize that set with the 'contained' field in
the nodes db, since it is easily possible for them to get out of sync.
(We can't require that the reverification queue table be in the same
database as the nodes table, so maintaining consistency with SQL
transactions is out. Plus, even if they were in the same database, using
such SQL transactions to maintain consistency would be slow and
unwieldy.)
This commit adds a method to the class representing the reverify queue
in the database, allowing us to get the list of every node that has at
least one record in the reverification queue.
Refs: https://github.com/storj/storj/issues/5431
Change-Id: Idce2633b3d63f2645170365e5cdeb2ea749fa9cb
This change implements the ranged loop observer to replace the audit
chore that builds the audit queue.
The strategy employed by this change is to use a collector for each
segment range to build separate per-node segment reservoirs that are
then merge them during the join step.
In previous observer migrations, there were only a handful of tests so
the strategy was to duplicate them. In this package, there are dozens
of tests that utilize the chore. To reduce code churn and maintenance
burden until the chore is removed, this change introduces a helper that
runs tests under both the chore and observer, providing a pair of
functions that can be used to pause or run the queueing function.
https://github.com/storj/storj/issues/5232
Change-Id: I8bb4b4e55cf98b1aac9f26307e3a9a355cb3f506
This change creates a new independent process, the 'auditor', comparable
to the repairer, gc, and api processes. This will allow auditors to be
scaled independently of the core.
Refs: https://github.com/storj/storj/issues/5251
Change-Id: I8a29eeb0a6e35753dfa0eab5c1246048065d1e91
Now that all the reverification changes have been made and the old code
is out of the way, this commit renames the new things back to the old
names. Mostly, this involves renaming "newContainment" to "containment"
or "NewContainment" to "Containment", but there are a few other renames
that have been promised and are carried out here.
Refs: https://github.com/storj/storj/issues/5230
Change-Id: I34e2b857ea338acbb8421cdac18b17f2974f233c
Now that we are doing scalable piecewise reverifications, the code for
handling the old way of doing things (containment, pending audits,
reporting, testing) can now be removed.
Refs: https://github.com/storj/storj/issues/5230
Change-Id: Ief1a75f423eff682e8f3d57804e343b3409a6631
This commit pulls the big switch! We have been setting up piecewise
reverifications (the workers for which can be scaled independently of
the core) for several commits now, and this commit actually begins
making use of them.
The core of this commit is fairly small, but it requires changing the
semantics in all the tests that relate to reverifications, so it ends up
being a large change. The changes to the tests are mostly mechanical and
repetitive, though, so reviewers needn't worry much.
Refs: https://github.com/storj/storj/issues/5230
Change-Id: Ibb421cc021664fd6e0096ffdf5b402a69b2d6f18
This change fixes the access of unset segments and keys on the reservoir
when the reservoir size is less than the max OR the number of sampled
segments is smaller than the reservoir size. It does so by tucking away
the segments and keys behind methods that return properly sized slices
into the segments/keys arrays.
It also fixes a bug in the housekeeping for the internal index variable
that holds onto how many items in the array have been populated. As part
of this fix, it changes the type of index to int8, which reduces the
size of the reservoir struct by 8 bytes.
The tests have been updated to provide better coverage for this case.
Change-Id: I3ceb17b692fe456fc4c1ca5d67d35c96aeb0a169
This method on the Verifier allows the caller to find, out of the nodes
holding pieces in a given segment, which ones are contained.
This method is not yet being used. It will be in a future commit.
Refs: https://github.com/storj/storj/issues/5230
Change-Id: I242cd999913ca4dabbe8a62767ed4869b31fca04
Here we add a worker class comparable to audit.Worker, which will be
responsible for pulling items off of the reverification queue and
calling reverifier.ReverifyPiece on them.
Note that piecewise reverification audits (which this will control) are
not yet being done. That is, nothing is being added to the
reverification queue at this point.
Refs: https://github.com/storj/storj/issues/5251
Change-Id: I94e28830e27caa49f2c8bd4a2336533e187ab69c
The Reporter is responsible for processing results from auditing
operations, logging the results, disqualifying nodes that reached
the maximum reverification count, and passing the results on to
the reputation system.
In this commit, we extend the Reporter so that it knows how to process
the results of piecewise reverification audits.
We also change most reporter-related tests so that reverifications
happen as piecewise reverification audits, exercising the new code.
Note that piecewise reverification audits are not yet being done outside
of tests. In a later commit, we will switch from doing segmentwise
reverifications to piecewise reverifications, as part of the
audit-scaling effort.
Refs: https://github.com/storj/storj/issues/5230
Change-Id: I9438164ce1ea4d9a1790d18d0e1046a8eb04d8e9
While researching logs from a large set of audits, I noticed that nearly
all of them had streamIDs starting with 0 or 1. This seemed very odd,
because streamIDs are supposed to be pretty much entirely random, and
every hex digit from 0-f should have been represented with roughly equal
frequency.
It turned out that our A-Chao implementation of reservoir sampling is
flawed. As far as we can tell, so is the Wikipedia implementation. No
one has yet reviewed the original 1982 paper by Dr. Chao in enough
detail to know where the error originated, but we do know that we have
been auditing segments near the beginning of the segment loop (low
streamIDs) far more often than segments near the end of the segment loop
(high streamIDs).
This change uses an algorithm Wikipedia calls "A-Res" instead, and adds
a test to check for that sort of bias creeping back in somehow. A-Res
will be slightly slower than A-Chao, because of a few extra steps that
need to be done, but it does appear to be selecting items uniformly.
Change-Id: I45eba4c522bafc729cebe2aab6f3fe65cd6336be
First, adding a logger argument allows the caller to have a logger
already set up with whatever extra fields they want here.
Secondly, we need to return the Outcome instead of a simple boolean so
that it can be passed on to the Reporter later (need to make the right
decision on increasing reputation vs decreasing it).
Thirdly, we collect the cached reputation information from the overlay
when creating the Orders, and return it from ReverifyPiece. This will
allow the Reporter to determine what reputation-status fields need to be
updated, similarly to how we include a map of ReputationStatus objects
in an audit.Report.
Refs: https://github.com/storj/storj/issues/5251
Change-Id: I5700b9ce543d18b857b81e684323b2d21c498cd8
NewContainment will replace Containment later in this commit chain, but
for now it is not yet being used.
NewContainment will allow a node to be contained for multiple pending
reverify jobs at a time. It is implemented by way of the reverify queue.
Refs: https://github.com/storj/storj/issues/5231
Change-Id: I126eda0b3dfc4710a88fe4a5f41780618ec19101
Adding a new worker comparable to Verifier, called Reverifier; as the
name suggests, it will be used for reverifications, whereas Verifier
will be used for verifications.
This allows distinct logging from the two classes, plus we can add some
configuration that is specific to the Reverifier.
There is a slight modification to GetNextJob that goes along with this.
This should have no impact on operational concerns.
Refs: https://github.com/storj/storj/issues/5251
Change-Id: Ie60d2d833bc5db8660bb463dd93c764bb40fc49c
audit.Queues was the previous method of passing stacks of segments for
audit to the verifier. As of commit 68f9ce4a, this is now happening
by way of the auditor queue (database-backed, so that communication can
happen between multiple peers). audit.Queues is no longer needed.
Refs: https://github.com/storj/storj/issues/5228
Change-Id: I46f2d48d655fb66366c92146cdb6b85aef200552
GetByNodeID will allow querying the reverification queue to see if there
are any pending jobs for a given node ID. And thus, to see if that node
ID should be contained or not.
Some parameters on the other methods of the ReverifyQueue interface have
been changed to accept pointers; this was done ahead of the rest of the
changes for the reverification queue to better match the signatures of
the methods that these will replace once ReverifyQueue is actually being
used (meaning fewer changes to tests).
Refs: https://github.com/storj/storj/issues/5251
Change-Id: Ic38ce6d2c650702b69f1c7244a224f00a34893a1
The audit chore will be pushing a large number of segments to be
audited, and the db might choke on that large insert when under load.
This change divides the insert up into batches, which can be sized
however is optimal for the backing database. It also arranges for
segments to be inserted in the order of the primary key, which helps
performance on some systems.
Refs: https://github.com/storj/storj/issues/5228
Change-Id: I941f580f690d681b80c86faf4abca2995e37135d
As part of the effort of splitting out the auditor workers to their own
process, we are transitioning the communication between the auditor
chore and the verification workers to a queue implemented in the
database, rather than the sequence of in-memory queues we used to use.
This logical database is safely partitionable from the rest of
satelliteDB.
Refs: https://github.com/storj/storj/issues/5251
Change-Id: I6cd31ac5265423271fbafe6127a86172c5cb53dc
Since the auditor will be moving to a different process from the
metainfo loop, we need a way of communicating which segments have
been chosen for audit. This queue will be that communication, for now.
Contrast this with the queue for _re_verifications in commit 9c67f62f.
Refs: https://github.com/storj/storj/issues/5251
Change-Id: I9a269c7ef21e6c5e9c6e5e1f3db298fe159a8a79
ReverifyPiece() is not currently hooked up to anything, but is planned
to take the place of audit.(*Verifier).Reverify().
ReverifyPiece() works by downloading one piece in its entirety, rather
than pulling an entire stripe across many nodes.
Change-Id: Ie2c680f4d3c3b65273a72466a3f9f55c115b0311
This table will be used as a queue for pieces that need to be reverified
(a regular audit timed out on the owning node, so now that node is
contained and we need to validate the piece before un-containing it).
Refs: https://github.com/storj/storj/issues/5228
Change-Id: I5dcd26b6adced8674cbd81884c1543a61ea9d4c8
Two things were done to optimize audit observer:
* monik call was removed as we have different way to track it
* no new allocation for audit.Segment struct inside observer
Benchmark against 'main':
name old time/op new time/op delta
RemoteSegment/Cockroach/multiple_segments-8 5.85µs ± 1% 0.74µs ± 4% -87.28% (p=0.008 n=5+5)
name old alloc/op new alloc/op delta
RemoteSegment/Cockroach/multiple_segments-8 2.72kB ± 0% 0.00kB ~ (p=0.079 n=4+5)
name old allocs/op new allocs/op delta
RemoteSegment/Cockroach/multiple_segments-8 50.0 ± 0% 0.0 -100.00% (p=0.008 n=5+5)
Change-Id: Ib973e48782bad4346eee1cd5aee77f0a50f69258
We have an alert on `not_enough_shares_for_audit` which fires too
frequently. Every time so far, it has been because of a network blip of
some nature on the satellite side.
Satellite operators are expected to have other means in place for
alerting on network problems and fixing them, so it's not necessary for
the audit framework to act in that way.
Instead, in this change, we add three new metrics,
`audit_not_enough_nodes_online`, `audit_not_enough_shares_acquired`, and
`audit_suspected_network_problem`. When an audit fails, and emits
`not_enough_shares_for_audit`, we will now determine whether it looks
like we are having network problems (most errors are connection
failures, possibly also some successful connections which subsequently
time out) or whether something else has happened.
After this is deployed, we can remove the alert on
`not_enough_shares_for_audit` and add new alerts on
`audit_not_enough_nodes_online` and `audit_not_enough_shares_acquired`.
`audit_suspected_network_problem` does not need an alert.
Refs: https://github.com/storj/storj/issues/4669
Change-Id: Ibb256bc19d2578904f71f5229111ac98e5212fcb
This structure is entirely unused within the audit module, and is only
used by repair code. Accordingly, this change moves the structure from
audit code to repair code.
Also, we take the opportunity here to rename the structure to something
less generic.
Refs: https://github.com/storj/storj/issues/4669
Change-Id: If85b37e08620cda1fde2afe98206293e02b5c36e
This is in response to community feedback that our existing reputation
calculation is too likely to disqualify storage nodes unfairly with
extreme swings up and down.
For details and analysis, please see the data_loss_vs_dq_chance_sim.py
tool, the "tuning reputation further.ipynb" Jupyter notebook in the
storj/datascience repository, and the discussion at
https://forum.storj.io/t/tuning-audit-scoring/14084
In brief: changing the lambda and initial-alpha parameters in this way
causes the swings in reputation to be smaller and less likely to put a
node past the disqualification threshold unfairly.
Note: this change will cause a one-time reset of all (non-disqualified)
node reputations, because the new initial alpha value of 1000 is
dramatically different, and the disqualification threshold is going to
be much higher.
Change-Id: Id6dc4ba8fde1be3db4255b72282207bab5491ca3
I don't know why the go people thought this was a good idea, because
this automatic reformatting is bound to do the wrong thing sometimes,
which is very annoying. But I don't see a way to turn it off, so best to
get this change out of the way.
Change-Id: Ib5dbbca6a6f6fc944d76c9b511b8c904f796e4f3
We made optimization for segment loop observers to avoid
heavy monkit initialization on each call. It was applied to very
often executed methods. Unfortunately we used wrong monkit
method to track function times. Instead mon.Task we used
mon.Func().
https://github.com/spacemonkeygo/monkit#how-it-works
Change-Id: I9ca454dbd828c6b43ba09ca75c341991d2fd73a8
The MinDownloadTimeout 950ms and delay of 1s were quiet close, possibly
causing flaky behavior in TestVerifierSlowDownload.
Change-Id: I4f6c1554a118b21427357642abe39986fd0af38d
The ApplyUpdates() method on the reputation.DB interface acts like the
similar Update() method, but can allow for applying the changes from
multiple audit events, instead of only one.
This will be necessary for the reputation write cache, which will batch
up changes to each node's reputation in order to flush them
periodically.
Refs: https://github.com/storj/storj/issues/4601
Change-Id: I44cc47767ea2d9423166bb8fed080c8a11182041
For nodes in excluded areas, we don't necessarily want to remove them
from the pointer, but we do want to increase the number of pieces in the
segment in case those excluded area nodes go down. To do that, we
increase the number of pieces repaired by the number of pieces in
excluded areas.
Change-Id: I0424f1bcd7e93f33eb3eeeec79dbada3b3ea1f3a
In addition to upgrading the storj.io/common library, this change
moves off the TCPConnector in favor of the HybridConnector per
the deprecation warning.
Change-Id: I7e7e1e7568e8b95e4a99ad9caa158a799e68e1e3
inconsistency
The original design had a flaw which can potentially cause discrepancy
for nodes reputation status between reputations table and nodes table.
In the event of a failure(network issue, db failure, satellite failure, etc.)
happens between update to reputations table and update to nodes table, data
can be out of sync.
This PR tries to fix above issue by passing through node's reputation from
the beginning of an audit/repair(this data is from nodes table) to the next
update in reputation service. If the updated reputation status from the service
is different from the existing node status, the service will try to update nodes
table. In the case of a failure, the service will be able to try update nodes
table again since it can see the discrepancy of the data. This will allow
both tables to be in-sync eventually.
Change-Id: Ic22130b4503a594b7177237b18f7e68305c2f122
Currently the slow db was sleeping for 1s and the timeout for audit was
1s. There's a slight chance that the timeout won't trigger on such a
small difference.
Increase the slow node sleep to 10x of the timeout.
Hopefully fixes#4268
Change-Id: Ifdab45141b3fc7c62bde11813dbc534b3255fe59
We don't use this column for anything. If you want to know if a node is
contained, you can check the pending_audits table.
Change-Id: I5671722a5fc6e1749d3a49e187a56556000ff941
We don't use this column for anything. If you want to know if a node is
contained, you can check the pending_audits table.
Change-Id: I8da1d8e01a2dcaff63c5067a7927b5451424ad04
Since we are sharing the reporting logic between repair and audit. We
need to remove metric reporting logic in reporter.
Change-Id: Ib87295ab19079329e7438327d785a7f5c21d3b21
Currently TextMaxVerifyCount flakes in some tests, try increasing the
sleep time to ensure that things are slow enough to trigger the error
condition.
Also pass ctx to all the funcs so we can handle sleep better.
Change-Id: I605b6ea8b14a0a66d81a605ce3251f57a1669c00
