From StatsdData.php

 79     public function getMessage($withMetric = true)
 80     {
 81         if (!$withMetric) {
 82             $result = sprintf('%s:%s', $this->getKey(), $this->getValue());
 83         } else {
*84             $result = sprintf('%s:%s|%s', $this->getKey(), $this->getValue(), $this->getMetric());
 85         }
 86

I'm guessing there's code somewhere that's wrongly letting some kind of string, indirectly based on user input, serve directly as the name of a statsd metric. Maybe an in_array check is failing somewhere.

This looks strange. The php code allows not to have string bigger than INT_MAX (https://github.com/php/php-src/blob/master/ext/standard/formatted_print.c#L100)
But the number is 65106 times (near at 2^16 = 65536) bigger than 2.147.483.648 (= 2^31). The sprintf gets three strings, which itself could not be bigger than INT_MAX making it worse

One fix is to use 64-bit php, another is to limit the key length in BufferingStatsdDataFactory::normalizeMetricKey to something useful? I would assume it is not coming from the value/metric, more from the key.

One unsane key build seems to be in https://gerrit.wikimedia.org/g/mediawiki/extensions/WikimediaEvents/+/ff7f55f570d4953ca387e72d65b98ddbe168f9cc/includes/WikimediaEventsHooks.php#671
but that is for dewiki only and not sure what WebRequest::getVal can return

Another one is https://gerrit.wikimedia.org/g/mediawiki/extensions/ContentTranslation/+/36529fd9af20b92aecad3d7356acb8e93d93d9f1/api/ApiContentTranslationPublish.php#114 also depends on what WebRequest::getVal can return.

	public function getAllowedParams() {
		return [ // …
			'to' => [
				ApiBase::PARAM_REQUIRED => true,
			],
			/* … */
	}

	protected function getCategories( array $params ) {
		/* … */
		MediaWikiServices::getInstance()->getStatsdDataFactory()
			->increment( 'cx.publish.highmt.' . $params['to'] );
		/* … */
		foreach ( $categories as $index => $category ) {
			$category = $this->removeApiCategoryNamespacePrefix( $category, $params['to'] );
			/* … */
	}

	private function removeApiCategoryNamespacePrefix( $category, $targetLanguage ) {
		$targetLanguage = MediaWikiServices::getInstance()->getLanguageFactory()->getLanguage( $targetLanguage );
		/* … */
	}

@Nikerabbit Could your team confirm that this is currently unvalidated and unnormalized, and if so, to patch it such that it either only logs stats for valid/canonical lang codes or normalises it to one?

(I don't know if this API wants to support non-canonical lang codes, they probably work today, so maybe that can continue; otherwise the the code for normalizing could be moved earlier such that requests are rejected if they are not matching the canonical form.)

Ack, thank you.

In T274507#6854125, @Umherirrender wrote:

This looks strange. The php code allows not to have string bigger than INT_MAX (https://github.com/php/php-src/blob/master/ext/standard/formatted_print.c#L100)
But the number is 65106 times (near at 2^16 = 65536) bigger than 2.147.483.648 (= 2^31). The sprintf gets three strings, which itself could not be bigger than INT_MAX making it worse

One fix is to use 64-bit php, another is to limit the key length in BufferingStatsdDataFactory::normalizeMetricKey to something useful? I would assume it is not coming from the value/metric, more from the key.

We should probably set a limit here for good measure.
AFAIK graphite can't store metrics over 4096 in length (due to unix path limitations), so we could add a pretty uncontroversial limit there I expect.
Just need to decide what to do in cases where the limit is reached.

One unsane key build seems to be in https://gerrit.wikimedia.org/g/mediawiki/extensions/WikimediaEvents/+/ff7f55f570d4953ca387e72d65b98ddbe168f9cc/includes/WikimediaEventsHooks.php#671
but that is for dewiki only and not sure what WebRequest::getVal can return

I'm checking internally to see if I can get rid of this entirely.
If I can not I'll add a much stricter limit to it. 20-40 chars or so for the input bit.

Change 674537 had a related patch set uploaded (by Addshore; owner: Addshore):
[mediawiki/extensions/WikimediaEvents@master] Remove statsd metric from onBeforeInitializeWMDECampaign hook

https://gerrit.wikimedia.org/r/674537

Change 674537 merged by jenkins-bot:
[mediawiki/extensions/WikimediaEvents@master] Remove statsd metric from onBeforeInitializeWMDECampaign hook

https://gerrit.wikimedia.org/r/674537

Change 674573 had a related patch set uploaded (by Lucas Werkmeister (WMDE); owner: Lucas Werkmeister (WMDE)):
[mediawiki/extensions/WikimediaEvents@master] Update outdated comment

https://gerrit.wikimedia.org/r/674573

Change 674573 merged by jenkins-bot:
[mediawiki/extensions/WikimediaEvents@master] Update outdated comment

https://gerrit.wikimedia.org/r/674573