Tablet throttler: be explicit about client app name, exempt some apps from checks and heartbeat renewals #13195

shlomi-noach · 2023-05-30T07:09:30Z

Description

An extension to #13177. This PR includes everything in #13177 and then adds more logic. I intentionally split the two apart so we can discuss the two changes separately.

This PR is an alternative approach to #13187. It has more changes, but I prefer the approach in this PR over #13187

#13177 says: "if the replica tablet's throttler is checked, then as result the primary tablet's throttler should request more heartbeats".

However, there are some clients that:

Engage with the throttler all the time
Do not really need the throttler's work
Thereby needlessly cause a scenario where the primary throttler leases heartbeats continuously and infinitely. Which works against the idea of "on-demand" heartbeats.

The particular clients are these three members of tabletserver:

schema tracker
binlog watcher
messager

For now, we consider all other clients as "need to use the throttler".

With this PR, we formalize the "app name": the name by which a client identifies itself to the throttler.

Generally speaking, this is an arbitrary string, and we've been using these names semi-formally thus far. Some names are "online-ddl", "vreplication", "gh-ost" etc. And command like ALTER VITESS_MIGRATION THROTTLE ALL tell the throttler to explicitly throttler calls made by specific apps ("online-ddl" in this example).

What's formalized in this PR is:

We move all names into a dedicated, single place, throttler/app.go. The names are constants, and now used throughout the code including endtoend tests.
We are much more granular with names. Again, names are arbitrary. For example, rowstreamer now uses the name "rowstreamer", and the external connector uses the name "external-connector".
Specifically, and critical to this PR, we break down uses of tabletserver's Engine's throttler. This throttler is used to VStream rows, and is shared by multiple clients. These particular three clients are of special interest:
- schema tracker
- binlog watcher
- messager
  What's so special about these three? When enabled, they're always running. Always checking. So these three, as others, clearly identify themselves by name when using the `tabletserver's Engine's throttler.
Next, the throttler now cares about the client's name. When asked to check, it first looks up the client app's name. Some names are now exempt from checks, and will not even cause heartbeat renewal. The three aforementioned apps are exactly those apps.

So what we get is:

based on Tablet throttler: non-PRIMARY tablets report back to PRIMARY throttler when they've been 'check'ed #13177, calls to the throttler, whether on PRIMARY or on REPLICA, ensure to renew the on-demand heartbeat lease
Except for particular apps, which the throttler basically always allows to go on, and for which it does not renew the heartbeat lease.

Related Issue(s)

Checklist

"Backport to:" labels have been added if this change should be back-ported
Tests were added or are not required
Did the new or modified tests pass consistently locally and on the CI
Documentation was added or is not required

Deployment Notes

…r when they've been 'check'ed Signed-off-by: Shlomi Noach <[email protected]>

Signed-off-by: Shlomi Noach <[email protected]>

…hey at all want to involve the throttler. Some lightweight clients, such as the schema tracker or the binlog watcher, or messager, do not need the throttler, and since some of these clients are _always on_, we also do not _want_ them to continuously approach the throttler. One side effect of always engaging with the throttler is the infinite renewal of on-demand heartbeats Signed-off-by: Shlomi Noach <[email protected]>

Signed-off-by: Shlomi Noach <[email protected]>

… in vstreamer.Engine. The throttler exempts specific apps from checks and will not renew heartbeats leases for those apps Signed-off-by: Shlomi Noach <[email protected]>

Signed-off-by: Shlomi Noach <[email protected]>

vitess-bot · 2023-05-30T07:09:33Z

Review Checklist

Hello reviewers! 👋 Please follow this checklist when reviewing this Pull Request.

General

Ensure that the Pull Request has a descriptive title.
If this is a change that users need to know about, please apply the release notes (needs details) label so that merging is blocked unless the summary release notes document is included.
If a test is added or modified, there should be a documentation on top of the test to explain what the expected behavior is what the test does.

If a new flag is being introduced:

Is it really necessary to add this flag?
Flag names should be clear and intuitive (as far as possible)
Help text should be descriptive.
Flag names should use dashes (-) as word separators rather than underscores (_).

If a workflow is added or modified:

Each item in Jobs should be named in order to mark it as required.
If the workflow should be required, the maintainer team should be notified.

Bug fixes

There should be at least one unit or end-to-end test.
The Pull Request description should include a link to an issue that describes the bug.

Non-trivial changes

There should be some code comments as to why things are implemented the way they are.

New/Existing features

Should be documented, either by modifying the existing documentation or creating new documentation.
New features should have a link to a feature request issue or an RFC that documents the use cases, corner cases and test cases.

Backward compatibility

Protobuf changes should be wire-compatible.
Changes to _vt tables and RPCs need to be backward compatible.
vtctl command output order should be stable and awk-able.
RPC changes should be compatible with vitess-operator
If a flag is removed, then it should also be removed from VTop, if used there.

mattlord

Makes sense to me. The existing tests should cover things.

I only had minor questions and comments. Let me know what you think. I can quickly review again and approve as needed.

mattlord · 2023-05-30T18:21:59Z

go/test/endtoend/onlineddl/vrepl/onlineddl_vrepl_test.go

@@ -400,11 +398,11 @@ func TestSchemaChange(t *testing.T) {
 		// to be strictly higher than started_timestamp
 		assert.GreaterOrEqual(t, lastThrottledTimestamp, startedTimestamp)
 		component := row.AsString("component_throttled", "")
-		assert.Contains(t, []string{string(vreplication.VCopierComponentName), string(vreplication.VPlayerComponentName)}, component)
+		assert.Contains(t, []string{throttlerapp.VCopierName, throttlerapp.VPlayerName}, component)


IMO it might be worth making a new type that's an alias for string so that what the string represents is even more explicit. This make it clear to future code writers that we're not using arbitrary strings.

You're right. I initially did that but then that affected so many lines of code, casting into and out of string, that I decided to hold off. I can still do that for this PR if you think it's worhtwhile.

I'll leave it up to you. I think it's probably worth it, but we can always do it later too.

Done in 37a7c2f. I'm not sure how I feel about it, what do you think?

mattlord · 2023-05-30T18:25:11Z

go/vt/vttablet/onlineddl/executor.go

@@ -1768,7 +1768,7 @@ export MYSQL_PWD
 		my ($self, %args) = @_;

 		return sub {
-			if (head("http://localhost:{{VTTABLET_PORT}}/throttler/check?app={{THROTTLER_ONLINE_DDL_APP}}:pt-osc:{{MIGRATION_UUID}}&p=low")) {
+			if (head("http://localhost:{{VTTABLET_PORT}}/throttler/check?app={{THROTTLER_ONLINE_DDL_APP}}:{{THROTTLER_PT_OSC_APP}}:{{MIGRATION_UUID}}&p=low")) {


I thought that we didn't support pt-osc? Might be a good opportunity to clean up the code if so.

It's labeled as "experimental". I'd rather not do everything in this PR. We can deprecate pt-osc later.

mattlord · 2023-05-30T18:29:23Z

go/vt/vttablet/tabletserver/throttle/throttler.go

+	// recentCheckTickerValue is an ever increasing number, incrementing once per second.
+	recentCheckTickerValue int64
+	// recentCheckValue is set to match or exceed recentCheckTickerValue whenever a "check" was made (other than by the throttler itself).
+	// when recentCheckValue < recentCheckTickerValue that means there hasn't been a recent check.


IMO 1 second ago is recent.

Agreed, and that's what the code considers as "recent". 1s. It's just that we have a 1 second granularity, and so the actual range can go as high as 1.999.

mattlord · 2023-05-30T18:35:06Z

go/vt/vttablet/tabletserver/throttle/throttler.go

+		if checkResult.RecentlyChecked {
+			// We have just probed a tablet, and it reported back that someone just recently "check"ed it.
+			// We therefore renew the heartbeats lease.
+			go throttler.heartbeatWriter.RequestHeartbeats()
+		}


My gut tells me there's the potential for a race here so that we don't request heartbeats for 1+ seconds which could throttle things and impact production traffic unnecessarily. Let's say that the throttler is checked once a second, but it's always just slightly behind the second tick.

Ah, but heartbeats are leased for some 5s-10s, and so there is no race. Leasing heartbeats to 1s is unadvisable. Now that you mention it, I should probably enforce a reasonable lower limit.

OK, makes sense. Thanks! Agreed on enforcing reasonable low and high limits on the on-demand heartbeat lease times.

mattlord · 2023-05-30T18:38:40Z

go/vt/vttablet/tabletserver/throttle/throttler.go

+		// This check was made by someone other than the throttler itself, i.e. this came from online-ddl or vreplication or other.
+		// We mark the fact that someone just made a check. If this is a REPLICA or RDONLY tables, this will be reported back
+		// to the PRIMARY so that it knows it must renew the heartbeat lease.
+		atomic.StoreInt64(&throttler.recentCheckValue, 1+atomic.LoadInt64(&throttler.recentCheckTickerValue))


Ah, so we add 1 to it. I think this at least largely addresses my earlier noted concern around timing.

mattlord · 2023-05-30T18:47:52Z

go/vt/vttablet/tabletserver/throttle/check_result.go

+	Threshold       float64 `json:"Threshold"`
+	Error           error   `json:"-"`
+	Message         string  `json:"Message"`
+	RecentlyChecked bool    `json:"RecentlyChecked"`


We have no synchronization around reading/writing this value, do we? If it's needed then we could make this an atomic.Bool.

This is not needed. We create a new CheckResult for each check.

Signed-off-by: Shlomi Noach <[email protected]>

mattlord

Thanks! ❤️

…io#13195 Signed-off-by: Shlomi Noach <[email protected]>

* Table throttler: --throttler-config-via-topo now defaults to 'true' Signed-off-by: Shlomi Noach <[email protected]> * add deprecation message Signed-off-by: Shlomi Noach <[email protected]> * endtoend tests: remove '--enable-lag-throttler' and use 'UpdateThrottlerConfig' everywhere Signed-off-by: Shlomi Noach <[email protected]> * always use vtctldclient Signed-off-by: Shlomi Noach <[email protected]> * use cluster.VtctldClientProcess Signed-off-by: Shlomi Noach <[email protected]> * disable --throttler-config-via-topo in old throttler tests Signed-off-by: Shlomi Noach <[email protected]> * Remove --throttler-config-via-topo where used, since it now defaults 'true' Signed-off-by: Shlomi Noach <[email protected]> * fix vreplication cluster setup, waiting for throttler config to apply Signed-off-by: Shlomi Noach <[email protected]> * changelog Signed-off-by: Shlomi Noach <[email protected]> * extend throttler threshold Signed-off-by: Shlomi Noach <[email protected]> * a bit more verbose Signed-off-by: Shlomi Noach <[email protected]> * fixed CLI test Signed-off-by: Shlomi Noach <[email protected]> * remove old '--enable-lag-throttler' flag, introduce '--heartbeat_on_demand_duration' Signed-off-by: Shlomi Noach <[email protected]> * more log info in throttler.Open() Signed-off-by: Shlomi Noach <[email protected]> * more logging Signed-off-by: Shlomi Noach <[email protected]> * Revert to --heartbeat_enable Signed-off-by: Shlomi Noach <[email protected]> * Protect throttler config change application with initMutex And in e2e test update the throttler config on the keyspace when it's created. Only wait for the new tablets in a shard to have the throttler enabled when adding a Shard. Signed-off-by: Matt Lord <[email protected]> * More CI testing Signed-off-by: Matt Lord <[email protected]> * CI testing cont Signed-off-by: Matt Lord <[email protected]> * Yes... Signed-off-by: Matt Lord <[email protected]> * Somebody doesn't like force pushes so msg here Signed-off-by: Matt Lord <[email protected]> * Increase on-demand heartbeat duration from 10s to 1m Signed-off-by: Matt Lord <[email protected]> * Use only on-demand heartbeats everywhere Signed-off-by: Matt Lord <[email protected]> * Use same throttler config everywhere Signed-off-by: Matt Lord <[email protected]> * Update all keyspaces and don't fail test on missing JSON keys Signed-off-by: Matt Lord <[email protected]> * Use constant heartbeats in vrepl e2e tests Until #13175 is fixed. Signed-off-by: Matt Lord <[email protected]> * Increase workflow command timeout Signed-off-by: Matt Lord <[email protected]> * Don't wait for throttler on non-serving primaries Signed-off-by: Matt Lord <[email protected]> * #13175 is fixed, therefore re-instating on-deman heartbeats Signed-off-by: Shlomi Noach <[email protected]> * Added ToC Signed-off-by: Shlomi Noach <[email protected]> * Tweak comment and kick CI Signed-off-by: Matt Lord <[email protected]> * Treat isOpen as the ready/running signal. Also align all initMutex usage. Signed-off-by: Matt Lord <[email protected]> * Re-adjust comment Signed-off-by: Matt Lord <[email protected]> * Adjust CheckIsReady() to match OnlineDDL's expectation/usage This was only using IsReady() before, now it's using IsOpen() and IsReady(). Signed-off-by: Matt Lord <[email protected]> * Get rid of log messages from SrvKeyspaceWatcher when no node/key Signed-off-by: Matt Lord <[email protected]> * More corrections/tweaks Signed-off-by: Matt Lord <[email protected]> * Use more convenient/clear new IsRunning function Signed-off-by: Matt Lord <[email protected]> * Revert "Use more convenient/clear new IsRunning function" This reverts commit 9aef276 as this change was not correct. Signed-off-by: Matt Lord <[email protected]> * Further fix correct use of IsOpen(), IsRunning(), IsEnabled() Signed-off-by: Shlomi Noach <[email protected]> * throttler.throttledApps cannot be nil Signed-off-by: Shlomi Noach <[email protected]> * Remove --enable_lag_throttler flag Signed-off-by: Shlomi Noach <[email protected]> * Deprecate --throttler_config_via_topo Signed-off-by: Shlomi Noach <[email protected]> * remove throttler mitigation code, as the problem was solved in #13195 Signed-off-by: Shlomi Noach <[email protected]> * deperecate throttler config flags Signed-off-by: Shlomi Noach <[email protected]> * Removed tabletmanager_throttler and tabletmanager_throttler_custom_config tests Signed-off-by: Shlomi Noach <[email protected]> * changelog Signed-off-by: Shlomi Noach <[email protected]> * remove EnableThrottler() call Signed-off-by: Shlomi Noach <[email protected]> * restore default value Signed-off-by: Shlomi Noach <[email protected]> * update threshold Signed-off-by: Shlomi Noach <[email protected]> * update flags desc Signed-off-by: Shlomi Noach <[email protected]> * using atomic.Bool Signed-off-by: Shlomi Noach <[email protected]> * Update changelog/18.0/18.0.0/summary.md Co-authored-by: Matt Lord <[email protected]> Signed-off-by: Shlomi Noach <[email protected]> * use MarkDeprecated Signed-off-by: Shlomi Noach <[email protected]> * do not expect flags in vttablet --help Signed-off-by: Shlomi Noach <[email protected]> * remove --throttler-config-via-topo from examples scripts Signed-off-by: Shlomi Noach <[email protected]> --------- Signed-off-by: Shlomi Noach <[email protected]> Signed-off-by: Matt Lord <[email protected]> Co-authored-by: Matt Lord <[email protected]>

shlomi-noach added 6 commits May 28, 2023 17:01

Tablet throttler: non-PRIMARY tablets report back to PRIMARY throttle…

78f98fc

…r when they've been 'check'ed Signed-off-by: Shlomi Noach <[email protected]>

inclusive language

379ec2b

Signed-off-by: Shlomi Noach <[email protected]>

add app name in test

94b6f03

Signed-off-by: Shlomi Noach <[email protected]>

Enforcing use of explicit names in all throttler checks, specifically…

d242965

… in vstreamer.Engine. The throttler exempts specific apps from checks and will not renew heartbeats leases for those apps Signed-off-by: Shlomi Noach <[email protected]>

new formal throttler app names placeholder

f978086

Signed-off-by: Shlomi Noach <[email protected]>

vitess-bot bot added NeedsDescriptionUpdate The description is not clear or comprehensive enough, and needs work NeedsIssue A linked issue is missing for this Pull Request NeedsWebsiteDocsUpdate What it says labels May 30, 2023

github-actions bot added this to the v17.0.0 milestone May 30, 2023

shlomi-noach mentioned this pull request May 30, 2023

Tablet throttler: throttler-config-via-topo defaults 'true', deprecation message for old flags #13130

Merged

4 tasks

shlomi-noach marked this pull request as ready for review May 30, 2023 17:38

shlomi-noach requested review from mattlord, rohit-nayak-ps, derekperkins, harshit-gangal, systay, deepthi and GuptaManan100 as code owners May 30, 2023 17:38

mattlord reviewed May 30, 2023

View reviewed changes

formalize throttlerapp.Name

37a7c2f

Signed-off-by: Shlomi Noach <[email protected]>

shlomi-noach mentioned this pull request May 31, 2023

vstreamer: support 'useThrottler' so that clients can choose whther they at all want to involve the throttler #13187

Closed

4 tasks

Refactor names in endtoend tests

3aab151

Signed-off-by: Shlomi Noach <[email protected]>

mattlord added Type: Enhancement Logical improvement (somewhere between a bug and feature) Component: TabletManager and removed NeedsDescriptionUpdate The description is not clear or comprehensive enough, and needs work NeedsIssue A linked issue is missing for this Pull Request labels May 31, 2023

mattlord approved these changes May 31, 2023

View reviewed changes

rohit-nayak-ps approved these changes May 31, 2023

View reviewed changes

shlomi-noach removed the NeedsWebsiteDocsUpdate What it says label Jun 1, 2023

shlomi-noach merged commit 374b94c into vitessio:main Jun 1, 2023

shlomi-noach deleted the throttler-replica-propagate-check-selective-throttling-by-name branch June 1, 2023 04:09

mattlord added Backport to: release-17.0 and removed Backport to: release-17.0 labels Jun 1, 2023

shlomi-noach added a commit to planetscale/vitess that referenced this pull request Jun 5, 2023

remove throttler mitigation code, as the problem was solved in vitess…

a085d55

…io#13195 Signed-off-by: Shlomi Noach <[email protected]>

shlomi-noach mentioned this pull request Jun 6, 2023

Deprecating and removing tablet throttler CLI flags and tests #13246

Merged

4 tasks

shlomi-noach mentioned this pull request Jun 18, 2023

Tablet Throttler: v18 features and objectives #13341

Closed

shlomi-noach mentioned this pull request Jul 30, 2023

Throttler: exempt apps via UpdateThrottlerConfig --throttle-app-exempt #13666

Merged

4 tasks

shlomi-noach mentioned this pull request Feb 14, 2024

Feature Request: add more visibility/stats in Tablet Throttler #15223

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Tablet throttler: be explicit about client app name, exempt some apps from checks and heartbeat renewals #13195

Tablet throttler: be explicit about client app name, exempt some apps from checks and heartbeat renewals #13195

shlomi-noach commented May 30, 2023

vitess-bot bot commented May 30, 2023

mattlord left a comment

mattlord May 30, 2023

shlomi-noach May 30, 2023 •

edited

Loading

mattlord May 30, 2023

shlomi-noach May 31, 2023

mattlord May 30, 2023

shlomi-noach May 30, 2023

mattlord May 30, 2023

shlomi-noach May 30, 2023

mattlord May 30, 2023

shlomi-noach May 30, 2023

mattlord May 30, 2023 •

edited

Loading

mattlord May 30, 2023

mattlord May 30, 2023 •

edited

Loading

shlomi-noach May 31, 2023

mattlord left a comment

Tablet throttler: be explicit about client app name, exempt some apps from checks and heartbeat renewals #13195

Tablet throttler: be explicit about client app name, exempt some apps from checks and heartbeat renewals #13195

Conversation

shlomi-noach commented May 30, 2023

Description

Related Issue(s)

Checklist

Deployment Notes

vitess-bot bot commented May 30, 2023

Review Checklist

General

If a new flag is being introduced:

If a workflow is added or modified:

Bug fixes

Non-trivial changes

New/Existing features

Backward compatibility

mattlord left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

shlomi-noach May 30, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mattlord May 30, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mattlord May 30, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mattlord left a comment

Choose a reason for hiding this comment

shlomi-noach May 30, 2023 •

edited

Loading

mattlord May 30, 2023 •

edited

Loading

mattlord May 30, 2023 •

edited

Loading