Fork me on GitHub
Maddy Finfrock, Split17:10:31

Good morning everyone! 👋 We're so excited to be here at DOES to talk about feature flagging and experimentation with you. Who here has used Feature Flags before?

👀 1
Lia Hanson, Split22:10:47

<!here> We know there's some amazing Happy Hours going on tonight. I'd like to attend a few of them myself! However, if you're doing a Sponsor Happy Hour Crawl, please stop by We've pulled them out from their caves and will be putting them in the hot spot (maybe lukewarm). Our door prizes will be charity gift cards, for the charity of your choice. So stop on by, bring questions and a smile (not lukewarm for that).

👍 2
Gaurav Handa22:10:50

I have a quick question, is it possible to automatically increase the traffic to new version of services using some strategy

Gaurav Handa22:10:52

Use case: I deploy a new version of my service and want to do an incremental rollout, but instead of someone doing it manually, we setup a strategy that automatically increments the %age of traffic or backs out if results are not good.

Gaurav Handa22:10:19

I meant, does feature flag tool support this?

Maddy Finfrock, Split22:10:32

@gauravhanda25 That's a great question! I'll let my colleague @scott.dedoes jump in to answer

Channing Benson22:10:00

Hello @gauravhanda25. Are you referring to a Multi-Armed Bandit like functionality?

Scott Dedoes22:10:24

Thanks @channing.benson! You found the channel before I could tag you!

Gaurav Handa22:10:22

We don't want our ops team to sit and monitor the new versions and then manually adjust traffic

Channing Benson22:10:48

So, at this time, we do not have any direct support for MAB. But it is a very hot topic for product direction. Are you more concerned with getting the most out of a treatment that increases a business metric or avoiding unforeseen problems like increased load time or error rates?

Gaurav Handa22:10:58

Also, one of the top goal is zero downtime deployments

Gaurav Handa22:10:13

where the system recovers or backs-out without any manual intervention

Channing Benson22:10:45

Because for the latter situation (wanting to mitigate blast radius due to an unintended negative consequence) we do have Feature Monitoring that can alert you when a feature causes a metric to degrade beyond a given threshold. Fixing the problem though would require manual intervention in response to the alert. (Again, automation of this is another top-of-mind direction for the product.)

Gaurav Handa22:10:49

So, it I have understood correctly, the dev team can deploy the code prod, that doesn't necessarily means it receives traffic.

Gaurav Handa22:10:03

Its just a deployment not release, the feature flags will govern when the code is live

Gaurav Handa22:10:26

and as alert can be setup on flag, any degradation will require manual intervention.

Gaurav Handa22:10:39

Next question, is how will monitor the application?

Gaurav Handa22:10:06

does it integrate with tools like dynatrace/splunk/sumologic

Channing Benson22:10:27

Metrics can be created ith any events you can send to Split. We have a RUM agent for basic JavaScript web page metrics (load time, dom load)

Gaurav Handa22:10:48

how about the backend services?

Channing Benson22:10:52

We have an integration with New Relic for Java code.

Channing Benson22:10:25

Further integrations with APM tools are under consideration

Gaurav Handa22:10:37

very cool. Thanks !

Gaurav Handa22:10:51

Also does it integrate well with Mobile apps ?

Channing Benson22:10:05

Yes. We have SDKs for iOS and Android

Scott Dedoes22:10:11

Thanks for answering these questions Channing! I requested connection with you on LinkedIn Gaurav in case you have further questions after the show or want more information. Feel free to reach out with any more questions and stop by the booth for raffle and swag!

Channing Benson22:10:30

@gauravhanda25: You're very welcome. Let us know if you have any further questions and thanks for your interest.

Gaurav Handa22:10:22

Thanks for prompt reply !

👍 2