High Availability set up for API Manager, CPI and EMS

[AlexeyP]

Hello,

We have a requirement for highly available and performant APIs. As a general pattern, all APIs are exposed via API manager which calls another API provider (would it be CPI or another backend system). API manager and CPI are in Neo. EMS, which is we're about to start using are in CF.

What we're currently finding that the level of "declared" availability is not good enough:

• Weekends are one of the busiest time for business and, as a consequence, for our systems so maintenance outages will adversely impact them
• We've already experienced an outage that would've lead to a significant systems outage had we had APIs that require HA deployed in APIM

What would be a recommended set up/patterns for dealing with these cases?

One way of dealing with this could be to implement EMS messaging to queue incoming requests in case backend systems are not available but EMS is not perfect either and suffers from outages too.

Another option is to move APIM and CPI to CF to achieve zero-downtime upgrades, but maintenance windows and outages are still there.

Third option would be to implement some sort of queueing in the consuming application using a different technology but this defies the purpose of having a single integration platform.

Other options coming to my mind are: possible failover to another region, have two API gateways (SAP and non-SAP) to deal with cases when SAP API manager is not available.

SAP hasn't been very forthcoming in discussing this so I wonder if anyone has worked on something like this before and could share their experience.

Thank you

[VijayKonam]

I echo your issues Alexy. Even after two years, SAP does not have confirmation on CPI being on multi AZ on CF!

[AlexeyP]

Hi Vijay, agree, it's not clearer. We're moving to CF from Neo so hopefully there will be fewer maintenance outages but there could still be technical issues with the platform (regardless of whether it's on SAP or IaaS side) so we need to understand how overall resilience can be improved (including implementation based on AZ).

[ElijahM]

Hi Alexey,

Thank you for raising up this issue - would be interesting to understand a little more your experience. As you mentioned the best approach is to go with Cloud Foundry for Multi-Availability Zones that enables near Zero Downtime Maintenance. Multi-Availability zones are available today for API Management, and CPI will soon be implementing the same pattern. With this pattern we have seen > 99.9% availability for API Management and expect that this will be the same experience with CPI when available.

Maintenance windows do exist, but SAP tries to keep them to a minimum, and is working through the AZ concept to move towards fewer and fewer maintenance windows with shorter increments. You can see already the change here https://support.sap.com/en/my-support/systems-installations/cac/maintenance-windows.html with shorter windows already declared for both standalone and Integration Suite (the larger suite containing API Development and Management, Cloud Integration and 3rd party connectivity all together).

While we see for > 99% of customers the increased and increasing SLA is good enough, there are a few customers who have higher needs and a few have gone ahead with boutique solutions like you suggested above utilizing Multi-Regional Deployments, to avoid outages. This is usually achieved with a Global Traffic Manager of some kind routing traffic based on customer architecture and availability.
We would not recommend utilizing a Non-SAP API Management as an SAP API Management backup, as you will lose all benefit from prepackaged integrations, intra-SAP pre-built integration capabilities, and so forth, increasing development time by a very large margin.

Regards,
Elijah