Until now, whenever audits were recorded we would try to delete
the node from containment just in case it exists. Since we now
want to treat segment repair downloads as audits, this would
erroneously remove nodes from containment, as repair does not go
through a Reverify step. With this changeset, (Batch)UpdateStats
will not remove nodes from containment. The Reverify method will
remove all necessary nodes from containment.
Change-Id: Iabc9496293076dccba32ddfa028e92580b26167f
Improve UpdateCheckIn on a contended row:
name old time/op new time/op delta
UpdateCheckInContended-100x-32 2.29s ±55% 0.17s ±61% -92.45% (p=0.008 n=5+5)
Change-Id: I053ab9f1cff136c306e5fb57f5e355cdc0269a8c
Currently we were duplicating code for AS OF SYSTEM TIME in several
places. This replaces the code with using a method on
dbutil.Implementation.
As a consequence it's more useful to use a shorter name for
implementation - 'impl' should be sufficiently clear in the context.
Similarly, using AsOfSystemInterval and AsOfSystemTime to distinguish
between the two modes is useful and slightly shorter without causing
confusion.
Change-Id: Idefe55528efa758b6176591017b6572a8d443e3d
When audits are being recorded, we automatically add some SQL to remove
the node from the pending audits table in case it exists. They are
removed from pending audits even if the node was offline for the audit.
This is not the correct behavior.
Add statement to record audit results in reverify tests to ensure no
more false positives.
Change-Id: I186ae68bc5e7962ef6c5defbebc1d95e63596a17
errs.Class should not contain "error" in the name, since that causes a
lot of stutter in the error logs. As an example a log line could end up
looking like:
ERROR node stats service error: satellitedbs error: node stats database error: no rows
Whereas something like:
ERROR nodestats service: satellitedbs: nodestatsdb: no rows
Would contain all the necessary information without the stutter.
Change-Id: I7b7cb7e592ebab4bcfadc1eef11122584d2b20e0
Initially we duplicated the code to avoid large scale changes to
the packages. Now we are past metainfo refactor we can remove the
duplication.
Change-Id: I9d0b2756cc6e2a2f4d576afa408a15273a7e1cef
When nodes check in for the very first time, if the satellite can't ping
them back, they are inserted into the nodes table with
last_contact_success of '0001-01-01 00:00:00+00'. If the stray nodes
chore runs before the node can fix their problem, they are DQd.
Solution: when DQing stray nodes, dont DQ where last_contact_success =
'0001-01-01 00:00:00+00'::timestamptz
Change-Id: I477a02d5ef85b2c930ed6b7d99a4d1995169bca8
instead of only generating invoices for nodes that had some
activity, we generate it for every node so that we can find
and pay terminal nodes that did not meet thresholds before
we recognized them as terminal.
Change-Id: Ibb3433e1b35f1ddcfbe292c034238c9fa1b66c44
This change introduces a new config flag,
--overlay.audit-history.offline-suspension-enabled,
to toggle suspending nodes for offline audits.
If the flag is set to true, nodes will be suspended if they meet the
requirements.
If the flag is false, nodes will not be suspended. If they are already
suspended and/or under review, these will be cleared.
Change-Id: Ibeba759c42d6e504f6b7598120d4fd4dab85ca74
Previously we would select a limited number of nodes for DQ in a
CTE and run the update on that set in a single transaction. This
could lead to locking on the table, so instead we select and update
in separate transactions.
Change-Id: I1e802c0845e829eeadcee4fa382f58462515fdb1
We would like to log Node IDs and last contact successes of nodes DQd
in this manner. We would also like to avoid returning an unbounded list
of items from the db. Therefore we change the query to select a limited
number of nodes that meet the DQ conditions and iterate until 0 rows are
returned. Each column of the query is already indexed.
Change-Id: Iaec2d9b56e7202b7c2028ba21750d40c8dd506ee
Full scope:
private/testplanet,satellite/{overlay,satellitedb}
Description:
In most cases, downtime tracking with audits will eventually lead
to DQ for nodes who are unresponsive. However, if a stray node has no
pieces, it will not be audited and will thus never be disqualified.
This chore will check for nodes who have not successfully been contacted
in some set time and DQ them.
There are some new flags for toggling DQ of stray nodes and the timeframes
for running the chore and how long nodes can go without contact.
Change-Id: Ic9d41fdbf214736798925e728245180fb3c55615
GetSuccessfulNodeNotCheckedInSince and GetOfflineNodesLimited are overlay methods
which were only used by the previous downtime tracking system which has been removed.
These methods should also be removed.
Change-Id: Idb829d742e1f987e095604423fff656fe581183e
* Separate audit history interface into its own file in the overlay
package
* Add overlay.AuditHistory struct so that internalpb.AuditHistory is
only used from within the database layer
* Add overlay.GetAuditHistory function for features that will require
access to detailed audit history information
* Do not return full audit history from UpdateAuditHistory - callers to
that function only need to know the online score and whether a full
tracking period has been completed
* Move audit history tests out of satellite/satellitedb, since they are
independent of database implementation
Change-Id: I35b0c4ac23bbaabd80624f8a9631c3cb1a1f33bd
Nodes which are offline_suspended will no longer be considered for new
uploads. The current threshold that enters a node into offline
suspension is 0.6. Disqualification for offline suspension is still
disabled.
Change-Id: I0da9abf47167dd5bf6bb21e0bc2186e003e38d1a
Query nodes table using AS OF SYSTEM TIME '-10s' (by default) when on CRDB to alleviate contention on the nodes table and minimize CRDB retries. Queries for standard uploads are already cached, and node lookups for graceful exit uploads has retry logic so it isn't necessary for the nodes returned to be current.
the immediate need is to be able to move the repair queue back out
of cockroach if we can't save it.
Change-Id: If26001a4e6804f6bb8713b4aee7e4fd6254dc326
With the new overlay.AuditOutcome type for offline audits, the
IsUp field is redundant. If AuditOutcome != AuditOffline, then
the node is online.
In addition to removing the field itself, other changes needed
to be made regarding the relationship between 'uptime' and 'audits'.
Previously, uptime and audit outcome were completely separated. For
example, it was possible to update a node's stats to give it a
successful/failed/unknown audit while simultaneously indicating that
the node was offline by setting IsUp to false. This is no longer possible
under this changeset. Some test which did this have been changed slightly
in order to pass.
Also add new benchmarks for UpdateStats and BatchUpdateStats with different
audit outcomes.
Change-Id: I998892d615850b1f138dc62f9b050f720ea0926b
- Remove flag for switching off offline audit reporting.
- Change the overlay method used from UpdateUptime to BatchUpdateStats, as this
is where the new online scoring is done.
- Add a new overlay.AuditOutcome type: AuditOffline. Since we now use the same
method to record offline audits as success, failure, and unknown, we need to
distinguish offline audits from the rest.
Change-Id: Iadcfe10cf13466fa1a1c2dc542db8994a6423355
Previously if a node did not have audit history data for each of the
windows over the tracking period, we would give them the benefit of
the doubt and set their score to 1. This was to prevent nodes from
being suspended right out the gate. We need a minimum amount of data
to evaluate them.
However, a node who is actually failing at being online will have no
idea until they have received enough audits and we suspend them.
Instead, we will always use their real score, but use a flag to determine
whether they are eligible for suspension/dq.
Change-Id: I382218f12e8770f95d4bcddcf101ef348940cadf
Additionally, this PR changes NewNodeFraction devDefault and testplanet config from 0.05 to 1.
This is because many tests relied on selecting nodes that were reputable based on audit and uptime
counts of 0, in effect, selecting new nodes as reputable ones.
However, since reputation is now indicated by a vetted_at db field that is explicitly set
rather than implied by audit and uptime counts, it would be more complicated to try to
update all of the nodes' reputations before selecting nodes for tests.
Now we just allow all test nodes to be new if needed.
Change-Id: Ib9531be77408662315b948fd029cee925ed2ca1d
Add online score used for the new audit history offline tracking system
to the nodes table. This allows us easy access to the node's online
score for the storagenode dashboard as well as for data analysis.
Change-Id: Ie99be1192e5236862a5b3dbed2e5ef03b9169410
When a node's audit history "online score" passes below a configured
threshold, the node goes into "offline suspension" mode and begins a
review period, where the operator is given an opportunity to bring their
node back online.
After the review period passes, offline suspension is turned off for the
node.
In the future, if a node still has a bad online score at the end of the
review period, it will be disqualified. This is disabled right now.
In the future, if a node is in offline suspension, it will be treated as
"unhealthy". Right now, there are no consequences for being in offline
suspension.
Minor changes:
* Moves AuditHistoryConfig out of UpdateStats/BatchUpdateStats args and
into UpdateRequest.
* Adds "now" argument to UpdateStats/BatchUpdateStats args for easy
testing.
* Changes formatting strings inside buildUpdateStatement to use specific
types.
Change-Id: I032b60298840fc16e6ef831da750f2d57619a397
What:
Use the github.com/jackc/pgx postgresql driver in place of
github.com/lib/pq.
Why:
github.com/lib/pq has some problems with error handling and context
cancellations (i.e. it might even issue queries or DML statements more
than once! see https://github.com/lib/pq/issues/939). The
github.com/jackx/pgx library appears not to have these problems, and
also appears to be better engineered and implemented (in particular, it
doesn't use "exceptions by panic"). It should also give us some
performance improvements in some cases, and even more so if we can use
it directly instead of going through the database/sql layer.
Change-Id: Ia696d220f340a097dee9550a312d37de14ed2044
We should not be sending any type of orders to nodes that have completed
graceful exit with the current satellite. In particular, we should not
be trying to audit them, because that would be silly.
Change-Id: Ie2153e5739914ab696feefcdef28545ed70f84e4
Since we increased the number of audit workers from 1 to 2, we need to make sure
concurrent updates do not trample each other. We can do this by serializing the
transactions.
Change-Id: If1b2f71cabe3c779c12ffa33c0c3271778ac3ae0
This ensures that rows are closed to avoid leaks.
Also verifies that Err() is called, to ensure that no
error is left behind.
Change-Id: Idd1bec9bf479f40021da67b2c80ce83033149469
In jaeger, it shows that this function gets called repetitively in
a single request. Most of the time, it's less than 1ms. Therefore, it
doesn't add much value in our trace but create noises.
Change-Id: I20234f36bbcf0fc22f91e5e1a5634c0cad577ed0
What: As soon as a node passes the vetting criteria (total_audit_count and total_uptime_count
are greater than the configured thresholds), we set vetted_at to the current timestamp.
Why: We may want to use this timestamp in future development to select new vs vetted nodes.
It also allows flexibility in node vetting experiments and allows for better metrics around
vetting times.
Please describe the tests: satellitedb_test: TestUpdateStats and TestBatchUpdateStats make sure vetted_at is set appropriately
Please describe the performance impact: This change does add extra logic to BatchUpdateStats and UpdateStats and
commits another variable to the db (vetted_at), but this should be negligible.
Change-Id: I3de804549b5f1bc359da4935bc859758ceac261d
