Atlas Search Index Not Build

Hi there,
Recently we have created Atlas Search indexes for the databases but the search function always return empty result. We used Atlas API to create the indexes, it responses okay meaning that indexes has been created but in fact they have never built successfully. Moreover there are no alerts indicating the search indexes built successfully.

We would like to use the Atlas UI Explorer to view the indexes build status but we can’t, our cluster has uptown 6000 databases, the Atlas UI Explorer always crashed when loading 6k databases. We try another way, i.e using Atlas API to get indexes but sadly, it don’t return “status” fields.

We run into a trouble that Atlas search indexes have been created but never built successfully. To give you more details, our cluster recently exceeded 75% memory usage. In addition, we noticed that, in our RTPP, the “SYS MEM” metric is too high

Screen Shot 2020-09-13 at 15.15.00

Are there any ways that “SYS MEM” affect the Atlas Search Index building ?
Thanks and regards

Hi @dattannguyen and welcome in the MongoDB Community :muscle:!

From what I’m reading here, it sounds like your cluster is paralysed because it’s begging for more RAM.

6K databases sounds like a lot but maybe they are really small so I can’t tell. How much data do you have in this cluster?

From what I see here, I guess you are running an M50 on GCP which has a recommended disk size of 160GB which sounds like a good ratio for the 30 GB of RAM. Do you have more than that?

MongoDB needs RAM for everything it does and creating indexes is, indeed, an extra burden that should be conducted outside of the peak hours if that’s possible.

Also, indexes should fit in RAM. Always.

Please make sure you have enough RAM for your indexes + working set + queries & workload. If indexes represent more than 15% of your RAM, I would say it’s time to add more RAM or I would start being more selective when creating indexes and I would make sure all the indexes that exists are really useful and not just a waste of RAM.

Cheers,
Maxime.