Now that we are trying to identify the root cause of the satellite load limitations (i.e. currently the satellite has a max ability of 400 rps for uploads and we need this to be higher), we are using the golang diagnostic tools to collect insight into what the bottlenecks are. We currently have a debug endpoint to gather some cpu and mem data, but it could be useful to have continuous profiling. GCP stackdriver has support for continuous profiling so lets set that up and see if it is helpful to gather more data.
This PR adds support for [GCP continuous profiler](https://cloud.google.com/profiler) which allows enabling continuous cpu/mem profiling and the stats are sent to stackdriver in google cloud console.
To enable the continuous profiling for a storj component, do the following:
- prereq: the workload must be running in GKE and have Stackdriver Profiling IAM role permissions
- provide the config flag `debug.profilename` in the config.yaml file for the workload (i.e. satellite api process, etc). The profilename should be the workload name, for example "satellite-api".
- once the above config flag is provided, the profiler will be initialized and profiling stats will automatically be sent to GCP project where the workload is running and viewable in the Stackdriver Profile page in the console
The current implementation assumes the workload is running in GKE, however if we find if useful we can add support to enable this from anywhere. But for simplicity, its configured this way assuming the main goal is to enable in production systems.
Change-Id: Ibf8ebe2df7bf06fdd4951ee6a1e48854dd36ad47
This new repair timeout (configured as TotalTimeout) will include both
the time to download pieces and the time to upload pieces, as well as
the time to pop the segment from the repair queue.
This is a move from Github PR #3645.
Change-Id: I47d618f57285845d8473fcd285f7d9be9b4318c8
This change adds two new tables to process orders as fast as we used
to but in an asynchronous manner and with hopefully less storage
usage. This should help scale on cockroach, but limits us to one
worker. It lays the groundwork for the order processing pipeline to
be queue rather than database driven.
For more details, see the added fast billing changes blueprint.
It also fixes the orders db so that all the timestamps that are
passed to columns that do not contain a time zone are converted to
UTC at the last possible opportunity, making it less likely to use
the APIs incorrectly. We really should migrate to include timezones
on all of our timestamp columns.
Change-Id: Ibfda8e7a3d5972b7798fb61b31ff56419c64ea35
rationale: if GC kills the satellite, it would be nice to make
it through a repair checker sweep first
Change-Id: Id56171dc8e13940cfb6481e36a910bad077a01ed
rate
Graceful exit is very slow at the moment. Over the last couple days we
increase the batch size on Stefans satellite to 1000 but as a side
effect the error rate was increased. With a batch size of 500 the error
rate looks stable.
This PR will increase the default to batch size to 300. Graceful exit
will still be painful slow but at least it will be a bit faster. At the
same time this PR also increases the number of errors we tolerate. We
don't want to DQ slow storage nodes just because they didn't finish all
300 transfers in time. We want to give them more retries.
Change-Id: I92e3f99e116d4988457d8b902a88e85ed1bcc1a7
Currently SNs report their free disk space once per hour. If a node
becomes full, it has to wait until the next contact cycle begins to
report; all the while receiving and failing upload requests. By increasing
the minimum required disk space, we can give the storage nodes more time
to report their space before the completely fill up. This change goes
hand-in-hand with another change we want to implement: trigger capacity
report on SN immediately upon falling below threshold.
Change-Id: I12f778286c6c3f582438b0e2949765ac43325e27
Control Panel allows to control different chores and services.
Currently this adds controlling of cycles.
Change-Id: I734f1676b2a0d883b8f5ba937e93c45ac1a9ce21
Allow rate limit project cache to expire so we can make project level rate limit changes without restarting the satellite process.
Change-Id: I159ea22edff5de7cbfcd13bfe70898dcef770e42
For the last few month we had no issues with order submission. I would
call it stable and now it is time to risk a lower expire time. This will
increase the database performance on the satellite and it will reduce
the delay for billing.
The long term goal is 6h but for that step we need to change graceful
exit first. At the moment storage nodes would get disuqlaified for not
transfering alle pieces in less than 6 hours.
Change-Id: I421a2c2421c5374c4e706e2338f1c2161fedc14c
With this change RS configuration will be set on satellite. Uplink with
get RS values with BeginObject request and will use it. For backward
compatibility and to avoid super large change redundancy scheme stored
with bucket is not touched. This can be done in future.
Change-Id: Ia5f76fc10c37e2c44e4f7b8754f28eafe1f97eff
Limits how many times metainfo APIs can be called per second by project ID. If limit is exceeded, the API will return Unauthorized/Too Many requests.
Limit per second and the size of the limiter cache per project are configurable, as well as whether the limiter is enabled.
Tests added/updated for the new rate_limit field in projects table.
Tests added for exceeding limits and disableing limiter.
Change-Id: Ic8ad102de3b690a475809d4f684156d5715f20fa
live accounting used to be a cache to store writes before they are picked up during
the tally iteration, after which the cache is cleared. This created a window in which
users could potentially exceed the storage limit. This PR refactors live accounting to
hold current estimations of space used per project. This should also reduce DB load
since we no longer need to query the satellite DB when checking space used for limiting.
The mechanism by which the new live accounting system works is as follows:
During the upload of any segment, the size of that segment is added to its respective
project total in live accounting. At the beginning of the tally iteration we record
the current values in live accounting as `initialLiveTotals`. At the end of the tally
iteration we again record the current totals in live accounting as `latestLiveTotals`.
The metainfo loop observer in tally allows us to get the project totals from what it
observed in metainfo DB which are stored in `tallyProjectTotals`. However, for any
particular segment uploaded during the metainfo loop, the observer may or may not
have seen it. Thus, we take half of the difference between `latestLiveTotals` and
`initialLiveTotals`, and add that to the total that was found during tally and set that
as the new live accounting total.
Initially, live accounting was storing the total stored amount across all nodes rather than
the segment size, which is inconsistent with how we record amounts stored in the project
accounting DB, so we have refactored live accounting to record segment size
Change-Id: Ie48bfdef453428fcdc180b2d781a69d58fd927fb
this commit introduces the reported_serials table. its purpose is
to allow for blind writes into it as nodes report in so that we have
minimal contention. in order to continue to accurately account for
used bandwidth, though, we cannot immediately add the settled amount.
if we did, we would have to give up on blind writes.
the table's primary key is structured precisely so that we can quickly
find expired orders and so that we maximally benefit from rocksdb
path prefix compression. we do this by rounding the expires at time
forward to the next day, effectively giving us storagenode petnames
for free. and since there's no secondary index or foreign key
constraints, this design should use significantly less space than
the current used_serials table while also reducing contention.
after inserting the orders into the table, we have a chore that
periodically consumes all of the expired orders in it and inserts
them into the existing rollups tables. this is as if we changed
the nodes to report as the order expired rather than as soon as
possible, so the belief in correctness of the refactor is higher.
since we are able to process large batches of orders (typically
a day's worth), we can use the code to maximally batch inserts into
the rollup tables to make inserts as friendly as possible to
cockroach.
Change-Id: I25d609ca2679b8331979184f16c6d46d4f74c1a6
With the new storage node downtime tracking feature, we need remove current uptime reputation configs: UptimeReputationAlpha, UptimeReputationBeta, and
UptimeReputationDQ. This is the first step of removing the uptime
reputation columns from satellitedb
Change-Id: Ie8fab13295dbf545e33aeda0c4306cda4ba54e36
We don't want slowloris nodes to be able to indefinitely block
up the satellite, so add a timeout. Some monitoring inspection
showed the largest success times being on the order of 30s, so
a 1min timeout should be sufficient to kill the misbehaving nodes.
Change-Id: I5e2c3480a15f6304e37262d0a4d30d07eae99bb3
As per discussed we decided to rate limit how fast we iterate through
the metainfo database in the metainfo loop. This puts in place a
mechanism for rate limiting and burst limiting if need be in the future.
The default for this rate limiting is still no limits so it stays the
same as our previous functionality.
Change-Id: I950f7192962b0e49f082d2c4284e2d52b0a925c7
Adds check to see if storage nodes are eligible to initiate
graceful exit, by checking their CreatedAt date and seeing if
their "age" is greater than the new config value:
NodeMinAgeInMonths
The default for this value is 6 months for now.
https://storjlabs.atlassian.net/browse/V3-3357
Change-Id: Ib807ab8987ddb5a38a27a83886490f73fe8c5816
* satellite/console: Add X-Frame-Options and Referrer-Policy security headers
* Update to use CSP instead of XFO and include tardigrade.io
* Make FrameAncestors a config option
* Update satellite-config lock
* Make help text for FrameAncestors better
* add overall failure percentage check and inactive time frame check before sending a response to sno
* update comment
* delete node from transfer queue if it has been inactive for too long
* fix linting error
* add test config value
* fix nil pointer
* add config value into testplanet
* add unit test for overall failure threshold
* move timeframe threshold to chore
* update protolock
* add chore test
* add per peiece failure count logic
* change config name from EndpointMaxFailures to MaxFailuresPerPiece
* address comments
* fix linting error
* add error handling for no row returned from progress table
* fix test for graceful exit chore on storagenode
* fix typo InActive -> Inactive
* improve readability for failure threshold calculation
* update config lock
* change error handling for GetProgress in graceful exit endpoint on the satellite side
* return proper rpc error in endpoint
* add check in chore test for checking finish timestamp and queue
