nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

ArcGIS Enterprise HA without a load balancer

JoeHershman

I find the documentation a bit confusing on this topic.

Is it possible to implement an HA Enterprise deployment using only Web Adapter and not a 3rd party load balancer? If so where is documentation for this

@JonathanQuinn

Personally, I do not see how Web Adapter is setup in front of two Portal machines.

Finde mehr Beiträge mit dem Tag

Accepted answers

All comments

LanceKirby2

@JonathanQuinn Thanks for your response! Additional documentation would be highly appreciated.

JonathanQuinn

It is possible to set up an HA deployment without a load balancer. The Web Adaptor acts as a load balancer; actively checking the health of the machines in the site its registered to, using simple a round-robin method to send requests to backend machines, and only sending requests to healthy machines. However, its a single point of failure which is generally why we suggest a load balancer in front of two web adaptors.

The Web Adaptor is really only necessary if you're using Integrated Windows Authentication, so if that's the case, you'd want to minimize single points of failure by setting up two. Whether they're behind a load balancer, or in a failover cluster where one Web Adaptor instance is active and the other standby in the event the active one goes down, is up to the organization and the resources available to it.

We can look into updating the documentation related to Windows Failover Cluster and similar approaches in place of a load balancer.

LanceKirby2

Geez that's is expensive! Thank you for helping me out. I guess my take away is yeah you can use web adaptors and they will route requests between each server if one is busy, but if one is handling a request and fails that request is just dead in the water and not routed to the other running server. If there was a third party load balancer in place, it would route requests just like web adaptor, but in the case where a request fails it would send that over to the running server. So to really have high availability you need a third party load balcncer.

I imagine this is a real issue with long running async requests, but for something like a query the user may not notice it as much. Would you agree with that?

JoeHershman

I am not an infrastructure guy. One client uses an F5, which I do believe is one of the top brands on the market, and not cheap.

We are in process of trying to do an approach without a load balancer and to use Windows Failover clustering as a less expensive approach. This is on-going effort.

LanceKirby2

Yes that lays it out clearly @JoeHershman! If I could ask a few more questions, are there any recommended load balancers by Esri for this kind of config? Could you share with me which one you use? Do you know of any blog or resource aside from the esri doc here that walks through the process of setting up a third party load balancer (would be great if it dealt with IIS)?

JoeHershman

@LanceKirby2 In an HA configuration you need a single Url (DNS entry) that directs to the Web Adapters and also is used to federate. This is the Url of the load balancer. A request sent to that Url then goes to the load balancer which will send it to a running Web Adapter. In the setup you describe each Web Adapter has it's own Url. For example, WA1, WA2, WA3. If WA1 goes down, the system is still up and running but the request sent to WA1 fails, it is not redirected to WA2. That is one of the purposes of the load balancer (really the prime purpose in an ArcGIS Server deployment), to have a single DNS entry that will go to an up and running Web Adapter

Hope that helps

LanceKirby2

Sorry @JoeHershman, can you clarify does it go to the one still running or doesn't go to the one still running?

JoeHershman

@LanceKirby2 the thing is in that configuration you Web Adapters will all have different Urls. So if something fails it goes not go to the one still running

LanceKirby2

Hey @JonathanQuinn sorry to be dense, but is it possible to just use the web adaptor component on each individual server, register those in the same site and achieve high availability? Or do you need a third party load balancer in front of two ArcGIS Servers that both have their own web adaptor and are registered together in one site?

JoeHershman

Thank you that clarifies things

DEWright_CA

Thanks for that clarification; I had seen in my firewalls the polling at the portal-server level but never thought to look for polling at the WA level also.

JonathanQuinn

Bit of clarification; the Web Adaptor is aware of the machines in the site as it polls for them every minute. Once it knows which machines are part of the site, it uses round-robin requests to send traffic to each node; it doesn't prioritize sending traffic to the primary.

In terms of the documentation, a single web adaptor, on a single IIS instance, is a single point of failure, which is why that pattern is not described in the high availability documentation. Load balancers are typically fault tolerant, and if you want IWA, you'd put two web adaptors behind the load balancer.

DEWright_CA

Yes, when you configure your web-adaptor it references which are the machines in the Cluster and knows which is the active-node; so when the active-node fails it's health check from the passive-node it takes over and the Web-Adaptor gets it's response from that node.

There is some lag and potential down-time between the roll-over; but based on the configuration it is not horrible.

Nimm an der Unterhaltung teil

Angemeldete Mitglieder können Beiträge verfassen, Updates folgen und mehr. Neu hier? Registriere ein kostenloses Konto.

Schnellzugriffe

Community Help
Find useful guides, FAQs, and documents to help you navigate and make the most of Esri Community.