Commit Graph

98 Commits

Author SHA1 Message Date
Nikolay
f255800da3
app/vminsert: returns back memory optimisation ()
Production workload shows that it's useful optimisation.

Channel based objects pool allows to handle irregural data ingestion
requests and make memory allocations more smooth.
It's improves sync.Pool efficiency, since objects from sync.Pool removed
after 2 GC cycles. With GOGC=30 value, GC runs significantly more often.

https://github.com/VictoriaMetrics/VictoriaMetrics/issues/6733

### Checklist

The following checks are **mandatory**:

- [ ] My change adheres [VictoriaMetrics contributing
guidelines](https://docs.victoriametrics.com/contributing/).

---------

Signed-off-by: f41gh7 <nik@victoriametrics.com>
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Co-authored-by: hagen1778 <roman@victoriametrics.com>
2024-08-13 10:49:09 -04:00
Aliaksandr Valialkin
353766061b
app/{vminsert,vmselect}: pass proper args to metrics.UnregisterSet() after a8356f3a26 2024-07-15 20:27:40 +02:00
Aliaksandr Valialkin
498fe1cfa5
app/vminsert/common: remove obsolete optimization for reducing memory usage for InsertCtx pool
This optimization is no longer needed according to benchmarks with ingestion rate.

This simplifies the code a bit.
2024-04-20 20:51:38 +02:00
Aliaksandr Valialkin
1d6ebe6404
app/vminsert/netstorage: rename storageNode.broken to storageNode.isBroken for the sake of consistency with storageNode.isReadOnly 2024-02-27 14:42:45 +02:00
Aliaksandr Valialkin
be3f5d1c64
app/vminsert/netstorage: log the error message when pending data wasn't sent to vmstorage nodes because they were unavailable at graceful shutdown 2024-02-27 14:15:15 +02:00
Aliaksandr Valialkin
05f56c411e
app/vminsert/netstorage: prevent from infinite loop when the re-routing is enabled and all the vmstorage nodes are unavailable 2024-02-25 03:24:51 +02:00
Aliaksandr Valialkin
63d635a5e4
app: consistently use atomic.* types instead of atomic.* functions
See ea9e2b19a5
2024-02-24 03:06:14 +02:00
Aliaksandr Valialkin
63a43331a3
docs/Cluster-VictoriaMetrics.md: document -disableReroutingOnUnavailable command-line flag
This is a follow-up for 88f0d1572e
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5713
2024-02-05 15:17:09 +02:00
Muxa1L
88f0d1572e
Add flag to vminsert to disable rerouting when some of storage nodes are unavailable ()
* Flag to disable rerouting from unavailable storage nodes

* Update netstorage.go

* Fix fmt for netstorage.go
2024-02-05 12:46:57 +00:00
Aliaksandr Valialkin
cd4f641d32
app/{vmstorage,vmselect}: disable vmstorage->vmselect RPC compression by default in order to improve query performance 2024-01-23 02:29:13 +02:00
Aliaksandr Valialkin
d52fd73f18
all: add up to 10% random jitter to the interval between periodic tasks performed by various components
This should smooth CPU and RAM usage spikes related to these periodic tasks,
by reducing the probability that multiple concurrent periodic tasks are performed at the same time.
2024-01-22 18:39:16 +02:00
Aliaksandr Valialkin
7d40506744
lib/prompb: change type of Label.Name and Label.Value from []byte to string
This makes it more consistent with lib/prompbmarshal.Label
2024-01-16 20:41:37 +02:00
Nikolay
7c19d01e9a
app/vminsert: properly close vmstorage connection ()
* app/vminsert: properly close vmstorage connection
previously vmstorage may stuck in broken state until vminsert restarts
since vmstorage was marked as read-only and connection was broken to it.
checkReadonly function never marked connection as broken
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4870

* wip

---------

Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
2023-09-01 17:56:41 +02:00
Nikolay
fbe2795670
app/vminsert: fixes readonly check ()
* app/vminsert: fixes readonly check
previously vminsert doesn't check readOnly state for vmstorage, since check was never performed for nil buffer
In this case every 30 second storage node loss readonly state and received some data.
It caused re-routing and possible slow down for ingestion
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4870

* wip

---------

Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
2023-08-30 16:24:24 +02:00
Aliaksandr Valialkin
19d61737c1
app/{vminsert,vmselect}: follow-up after 2b7b3293c1
- Document the change at docs/CHANGELOG.md
- Set the default value for -vmstorageUserTimeout to 3 seconds. This is much better
  than the 0 value, which means that TCP connection to unreachable vmstorage could block
  for up to 16 minutes.
- Document -vmstorageUserTimeout at docs/Cluster-VictoriaMetrics.md
2023-08-29 12:17:39 +02:00
Will Jordan
2b7b3293c1
Add vmstorageUserTimeout flags to configure TCP user timeout (Linux) ()
`TCP_USER_TIMEOUT` (since Linux 2.6.37) specifies the maximum amount of
time that transmitted data may remain unacknowledged before TCP will
forcibly close the connection and return `ETIMEDOUT` to the application.

Setting a low TCP user timeout allows RPC connections quickly reroute
around unavailable storage nodes during network interruptions.
2023-08-29 11:46:39 +02:00
Nikolay
b0977b07fb
app/vminsert: adds note for dropSamplesOnOverload flag ()
Adds note for dropSamplesOnOverload flag that are samples dropped before replication
2023-08-10 12:18:29 +02:00
Aliaksandr Valialkin
992c300ce9
all: replace atomic.Value with atomic.Pointer[T]
This eliminates the need in .(*T) casting for results obtained from Load()

Leave atomic.Value for map, since atomic.Pointer[map[...]...] makes double pointer to map,
because map is already a pointer type.
2023-07-19 17:48:26 -07:00
Nikolay
5514b5d552
app/vminsert: allow parsing tenant id from ()
VictoriaMetrics_ProjectID and VictoriaMetrics_AccountID labels.
It should help to migrate for new labels vm_account_id vm_project_id without service downtime
2023-05-16 08:16:37 -07:00
Nikolay
82e2f19bc2
app/vminsert: correctly allocate buffer for storagenodes ()
in case of dynamic discovery number of nodes may change and we have to allocate new buffer for this case
otherwise vminsert may panic
2023-05-08 08:57:15 -07:00
Aliaksandr Valialkin
3a25a4b1de
app/{vminsert,vmselect}: speed up TestInitStopNodes() 2022-12-03 23:53:14 -08:00
Aliaksandr Valialkin
fe8d40f12c
app/{vminsert,vmselect}: test initialization with different number of storage nodes
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/3329
2022-11-09 11:48:39 +02:00
Aliaksandr Valialkin
8540dd669b
app/vminsert/netstorage: move nodesHash from global state to storageNodesBucket
This should prevent from panics when the list of discovered vmstorage nodes changes.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/3329
2022-11-09 11:45:24 +02:00
Aliaksandr Valialkin
976bbe3677
app/{vminsert,vmselect}: limit the access to storageNodes to getStorageNodesBucket and setStorageNodesBucket functions
This makes the code more maintainable and earier to test.
2022-10-28 11:41:55 +03:00
Aliaksandr Valialkin
4f53147ed4
app/{vminsert,vmselect}/netstorage: allow calling Init()+MustStop() in a loop
Previously netstorage.MustStop() call didn't free up all the resources,
so the subsequent call to nestorage.Init() would panic.

This allows writing tests, which call nestorage.Init() + nestorage.MustStop() in a loop.
2022-10-25 14:43:05 +03:00
Nikolay
505d359b39
app/vminsert: allows parsing tenant id from labels ()
* app/vminsert: allows parsing tenant id from labels
it should help mitigate issues with vmagent's multiTenant mode, which works incorrectly at heavy load
and it cannot handle more then 100 different tenants.
This functional hidden with flag and do not change vminsert default behaviour
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/2970

* Update docs/Cluster-VictoriaMetrics.md

Co-authored-by: Roman Khavronenko <roman@victoriametrics.com>

* wip

* app/vminsert/netstorage: clean remaining labels in order to free up GC

* docs/Cluster-VictoriaMetrics.md: typo fix

* wip

* wip

Co-authored-by: Roman Khavronenko <roman@victoriametrics.com>
Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
2022-09-30 17:28:35 +03:00
Aliaksandr Valialkin
dc58cb3fd8
app/vminsert/netstorage: pre-initialize the remaining throttled loggers
This is a follow-up after 6c66804fd3
2022-08-04 18:34:42 +03:00
Aliaksandr Valialkin
6c66804fd3
all: locate throttled loggers via logger.WithThrottler() only once and then use them
This reduces the contention on logThrottlerRegistryMu mutex when logger.WithThrottler()
is called frequently from concurrent goroutines.
2022-06-27 12:34:30 +03:00
Aliaksandr Valialkin
dceca7e864
all: remove explicit "xxhash" name when importing github.com/cespare/xxhash/v2 package
This is a follow-up for fe2269b999
2022-06-21 20:27:30 +03:00
Aliaksandr Valialkin
b28c6febf9
app/{vminsert,vmselect}: add -vmstorageDialTimeout command-line flag for tuning the maximum time needed for establishing connections to vmstorage
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/711
2022-06-20 15:17:34 +03:00
Aliaksandr Valialkin
e613ca9ba8
app/vminsert/netstorage: re-route samples from readonly vmstorage nodes to healthy nodes if -dropSamplesOnOverload command-line flag is set 2022-05-07 01:40:02 +03:00
Aliaksandr Valialkin
925fa9a7de
docs/Cluster-VictoriaMetrics.md: make the description for -rpc.disableCompression command-line flag more clear 2022-05-06 16:24:56 +03:00
Aliaksandr Valialkin
8752cce157
app/vminsert: reduce the max packet size, which vminsert can send to vmstorage
This reduces the max memory usage for vminsert and vmstorage under heavy ingestion rate
by up to 50% on production workload
2022-04-05 15:39:58 +03:00
Aliaksandr Valialkin
755e26e67b
app/vminsert/netstorage: throttle warning logs, which can be too verbose when vminsert cannot send data to vmstorage nodes 2022-02-07 15:39:06 +02:00
Aliaksandr Valialkin
9fb5ce5fb6
app/vminsert/netstorage: log vmstorage addr, which cannot accept new samples due to overload and/or unavailability 2022-02-07 15:03:05 +02:00
Aliaksandr Valialkin
5aee6eb406
app/vminsert/netstorage: tune re-routing algorithm further 2022-02-07 14:35:39 +02:00
Aliaksandr Valialkin
1a5546006d
app/vminsert/netstorage: typo fix after 021ee53ba8 2022-02-07 13:20:43 +02:00
Aliaksandr Valialkin
c738739494
app/vminsert: add -dropSamplesOnOverload command-line flag
Drop incoming samples if the destination vmstorage node is unavailable
and/or accepts data at slower rate than other vmstorage nodes
2022-02-07 12:32:18 +02:00
Aliaksandr Valialkin
021ee53ba8
app/vminsert: improve re-routing logic in order to spread rows more evenly among the available storage nodes 2022-02-06 20:20:02 +02:00
Aliaksandr Valialkin
4fddcf4c83
app/{vminsert,vmstorage}: follow-up after a171916ef5
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/269
2021-10-08 14:09:51 +03:00
Nikolay
a171916ef5
Adds read-only mode for vmstorage node ()
* adds read-only mode for vmstorage
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/269

* changes order a bit

* moves isFreeDiskLimitReached var to storage struct
renames functions to be consistent
change protoparser api - with optional storage limit check for given openned storage

* renames freeSpaceLimit to ReadOnly
2021-10-08 12:52:56 +03:00
Aliaksandr Valialkin
64b6f3f1c8
app/vminsert: fix uneven distribution of time series among storage nodes
Use distinct seed for distribution hash calculations on the second level of vminsert nodes.

See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1672
2021-10-07 12:22:39 +03:00
Aliaksandr Valialkin
7ad54041fe app/{vminsert,vmselect}: automatically add missing port in -storageNode lists passed to vminsert and vmselect
This should simplify manual setup of the cluster according to https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#cluster-setup
2021-09-15 18:04:30 +03:00
Aliaksandr Valialkin
a1b298a842 app/vminsert/netstorage: disable rerouting by default
Production clusters work more stable with the disabled rerouting during rolling restarts and/or
during spikes in time series churn rate. So it would be better disabling the rerouting by default.

The re-routing can be enabled by passing `-disableRerouting=false` command-line flag to `vminsert` nodes.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/791
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1054
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1165
2021-09-13 18:50:47 +03:00
Aliaksandr Valialkin
dca6f0f7de app/vminsert/netstorage: remove the limit on the number of -storageNode addresses
There is no any reasons to limit the number of `-storageNode` addresses to 255.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1599
2021-09-09 19:11:36 +03:00
Aliaksandr Valialkin
9eb828b2c2 app/vminsert: add vm_rpc_send_duration_seconds_total metric per each vminsert->vmstorage link
This metric is useful for determining high link saturation with the following alerting rule:

rate(vm_rpc_send_duration_seconds_total) > 0.9s
2021-08-11 11:42:33 +03:00
Aliaksandr Valialkin
c18017a9c3 app/vminsert/netstorage: sort the -storageNode list passed to vminsert nodes
This should reduce resource usage (CPU, RAM, disk IO) at vmstorage nodes
if the addresses of vmstorage nodes are passed in random order to vminsert nodes.
2021-06-23 14:00:08 +03:00
Aliaksandr Valialkin
2c6b917749 app/vminsert/netstorage: update storageNode.lastRerouteTime before the rerouting
This is needed for reliable detection of storage nodes with recent rerouting
2021-06-08 12:06:32 +03:00
Aliaksandr Valialkin
0d067eb112 app/vminsert/netstorage: tune re-routing algorithm
Do not re-route data to unavailable storage node. Send it to the remaining storage nodes instead
even if they cannot keep up with the load. This should spread the load more evenly among available
storage nodes.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/791
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1054
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1165
2021-06-05 16:23:44 +03:00
Aliaksandr Valialkin
1c09e71f5b app/vminsert: add -disableRerouting command-line flag for disabling re-routing if some vmstorage nodes have lower performance than the others
Refactor the rerouting mechanism and make it more resilient to cases when some of vmstorage nodes are temporarily unavailable.

Reduce the probability of rerouting storm.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/791
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1054
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1165
2021-06-04 04:33:52 +03:00