When pieces fail an audit (hard fail, meaning the node acknowledged it
did not have the piece or the piece was corrupted), we will now remove
those pieces from the segment.
Previously, we did not do this, and some node operators were seeing the
same missing piece audited over and over again and losing reputation
every time.
This change will include both verification and reverification audits. It
will also apply to pieces found to be bad during repair, if
repair-to-reputation reporting is enabled.
Change-Id: I0ca7af7e3fecdc0aebbd34fee4be3a0eab53f4f7
If a node is already disqualified or exited, don't bother going through
the audit reputation code. If there's an additional reputation update to
be made based on the automatic offline audit these nodes would get, the
overlay DB rejects it anyway.
Change-Id: I48ba151aa640b32f87e1285521be465d1f692a78
This change modifies "reputation change trigger" logic for dq and
suspension in order to:
* count only nil -> not-nil and not-nil -> nil as "reputation changes"
* do not count the timestamp updating (not-nil -> not-nil) as a
reputation change
Change-Id: Ie57c97e80e5f3b368e2f2dc004e680a2f9416d1c
`overlay.(*Service).UpdateReputation()` takes a "reputationChanges"
parameter, a slice of node events indicating whether we think the node's
disqualification or suspension status is changing. This is necessary so
that the overlay service can notify the nodeevents DB about these
changes.
In several cases, however, this list of events is not constructed
correctly, because of missing information about the previous state.
In most cases, this is because the node was offline, and the order limit
creation functions (which usually obtain and return the prior reputation
status) ignored that node.
This change makes it so that all callers to
`overlay.(*Service).UpdateReputation()` can be expected to provide a
correct list of change events (as correct as feasible, given that we
can't lock the node's information in the database during the entire
operation).
It ended up that there was only one caller we needed to worry about, and
that was reputation.(*Service).ApplyAudit(). So the bulk of this change
is teaching that function how to recognize when the prior reputation
status was not filled in, and fill it in.
Refs: https://github.com/storj/storj/issues/5464
Change-Id: I52ce385fc9c0ce3b283b998d517998e7f4ec8792
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
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
The procedure responsible for node reputation status comparison could
return an invalid result due to comparing a status timestamp against
itself rather than comparing it against another. This results in
unnecessary database updates that could be avoided otherwise.
This change modifies the procedure to resolve this issue.
Change-Id: Id147e1942e994e8bca4ced2a9358f2474927d6ec
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
In an effort to distribute load on the reputation database, the
reputation write cache scheduled nodes to be written at a time offset by
the local nodeID. The idea was that no two repair workers would have the
same nodeID, so they would not tend to write to the same row at the same
time.
Instead, since all satellite processes share the same satellite ID
(duh), this caused _all_ workers to try and write to the same row at the
same time _always_. This was not ideal.
This change uses a random number instead of the satellite ID. The random
number is sourced from the number of nanoseconds since the Unix epoch.
As long as workers are not started at the exact same nanosecond, they
ought to get well-distributed offsets.
Change-Id: I149bdaa6ca1ee6043cfedcf1489dd9d3e3c7a163
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
This has been a cause of some confusion, even though the fields are
labeled as being copies of config values.
Having them be under a field explicitly named "Config" makes this
clearer, plus, allows the values to be passed in simply as a copy
of the Config struct from the satellite, rather than copying the fields
individually (which can be error-prone, particularly as the AuditCount
field in UpdateRequest is apparently not the same thing as the
AuditCount field in reputation.Config).
Refs: https://github.com/storj/storj/issues/4601
Change-Id: I386953347b71068596618616934aa28e3245cdc1
The two protobuf types are identical except that one is in our common/pb
package, and the other is in internalpb. Since the type is public
already, and there is no difference in the internal one, it seems better
to use the public one for all satellite needs.
There is also another type which is essentially identical, but which is
not a protobuf type, also called "AuditHistory". It looks like we don't
ever actually need to have a separate type from the protobuf one.
This change makes us use "storj/common/pb".AuditHistory for all of our
AuditHistory needs.
Refs: https://github.com/storj/storj/issues/4601
Change-Id: If845fde21bb31c801db6d67ffc9a146d1617b991
This functionality will be needed in both packages, so here we move it
into the more general reputation-code package and export it for use in
satellitedb.
This also removes the related UpdateAuditHistory() signature from the
reputation DB interface, since it doesn't have anything to do with the
db. It doesn't need to be a method, either.
Finally, this changes the test for addAudit to be a plain test function
instead of using testplanet.Run(). It didn't need a whole testplanet
setup or any databases.
Refs: https://github.com/storj/storj/issues/4601
Change-Id: I90f6a909e5404f03ad776b95cfa2f248308c57c1
This will let us update our reputation cache when writing through to the
db.
Since the information is already being fetched from the db and returned
to the application, the extra cpu load here should be minimal.
Refs: https://github.com/storj/storj/issues/4601
Change-Id: I2b8619f2c0d541893c7d3e7d33b1863b96775ebd
We added nodes.disqualification_reason recently, but we didn't add a
corresponding column in the reputations table (despite having a
corresponding `disqualified` column there).
Without this change, the (very useful and informative) assignments to
updateFields.DisqualificationReason in reputations.go have no effect.
Refs: https://github.com/storj/storj/issues/4601
Change-Id: I77404902ca64b56aed72f1de76b303fe82b76aab
Set disqualification reason when reputations stats are updated on DB.Update.
Added tests for DisqualifyNode and for disqualification cases which happens during Update.
Change-Id: I00130ab5d9722422805159ad2f183c205de60f7e
Remove redundant suspension timestamp column from nodes and reputation tables.
Suspended timestamp was moved to unknown_audit_suspended and suspended column is
no longer used so there is no point in keeping both.
Change-Id: Ieea3f12141b33ec9efe7594f4c9dbc7e10675b0e
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
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
gracefulexit
With the effort to move audit related data into reputation store, this
PR updates gracefulexit endpoint to use reputation service to get a
node's audit score
Change-Id: Iad93ea689ad67ff9c57c7be16687e21e715fab7a
Currently, reputation table is only populated when a node has been
audited. This is ok in production, however a lot of our tests doesn't
upload any data or trigger audits.
This PR adds an initialization step in testplanet to populate reputation
table with zero value for nodes reputation.
Change-Id: I11b381236669db346dc68a48a6d4a27334a0a8b8
package in audit
This PR implements reputation store and replace overlay in audit service
to use such store for storing node's audit stats.
In order to keep the changeset smaller, most of the changes in this PR is for copying audit logic in overlay to
reputation package. In a following PR, the duplicating code will be
removed from overlay.
Change-Id: I16c12494a0970f44c422b26cf603c1dc489e5bc1
Define service and DB interface for storing node reputation data
and updating the overlay cache.
Add overlay service and DB method UpdateReputation.
See https://github.com/storj/storj/pull/4144
Change-Id: Iedd8bd3274457d26c595919303d55327c1464b8c
