Zero downtime
-
Hello there! We are looking for some free identities providers and some names crossed our minds such as fusionauth, keycloak, auth0, zitadel, etc.
Currently we use WSO2 with a very outdated version and we have had a huge headache trying to update it in the recent past. Even with the paid help from the WSO2 team, it failed.
The most important requirement is to perform well with a load of millions of users. We currently have around 7K tenants and over 8.5M users (although monthly active is about 600k). At peak hours it reaches 1.8~2K login requests per minute.
We use AWS, so we tested fusionauth as a deployment on Amazon EKS and an RDS instance with postgres. We haven't tested ElasticSearch as a search engine though. Scaling it horizontally and setting higher JVM memory under the fusionauth-app.memory environment variable managed to handled the 7k tenants and 10.5 millions users in database with loads of 50 requests per second (3k per minute).
That said, what concerns us most is the limitation we found here: https://fusionauth.io/docs/v1/tech/reference/limitations#system-upgrade-downtime (open issue: https://github.com/FusionAuth/fusionauth-issues/issues/1240).
If I understand correctly, every time the database needs to be migrated it will cost us some donwtime. If yes, is there a prediction to solve it to be possible doing the migration with zero downtime? For us, donwtime is 100% out of the question, even for a very short period of time. And we don't want to fall into the same trap that was with WSO2, freezing its version and being very outdated.
Thanks!
-
@alison-rafaelc Hi Alison,
It depends on how you architect the FusionAuth system, but in our FusionAuth cloud system, we see downtime of seconds to minutes for system upgrades (depending on the number of users and the data being migrated). We get this by swapping out one node at a time, running in an 3+ node cluster. Some details here: https://fusionauth.io/docs/v1/tech/installation-guide/cloud#upgrade-duration
Worth noting that you control when you do the upgrade, unlike a typical multi-tenant saas.
We have a number of customers and community members with thousands of tenants and millions of users so the numbers you mention seem fine. I'm glad you were able to stand up a FusionAuth cluster and load test it.
Zero downtime upgrades are something we've discussed internally and have a strategy for, it just hasn't made it onto the roadmap yet.
If you'd like to have a discussion with a technical sales team about FusionAuth and SLAs, I'm happy to have someone reach out to you.
