Out Of Memory Error with Full Ejection?


#1

Hello all,

I have the following warning on my cluster:

Hard Out Of Memory Error. Bucket “parse3” on node cbcluster2 is full. All memory allocated to this bucket is used for metadata. (repeated 2 times)

I am using 4.5, my bucket named parse3 RAM quota = 1000MB and Full Ejection.

I have enough CPU and Disk I/O’s to sustain activity on this cluster with 3 nodes. But I don’t have the luxury of having more nodes and more than 14.2 GIG of RAM on each Nodes.
I will eventually have dozens of buckets to manage. This is a Development environment. As long as performance is adequate, I wont have more CPU/RAM/disk speed.

I thought by putting buckets to Full Ejection would prevent me from having those warning messages. Is there something else I have to do?

Thanks,

Steeve


#2

That should work. Just test your rebalance performance to make sure it doesn’t take hours to complete. Test with something close to your production size in the bucket

You really don’t want to do that. Try name spacing your keys EX. 1234_key , 5678_key and putting them in as few buckets as possible, like 1-5 buckets. As each bucket will create 1024 data files X 2 (for replica of data files) + logs and metrics for each bucket.


#3

Our application fails because of the memory error:
Operation(s) failed: [0x8] The server is out of memory. Try again later
Operation(s) failed: [0x8] The server is out of memory. Try again later
Operation(s) failed: [0x8] The server is out of memory. Try again later
Operation(s) failed: [0x8] The server is out of memory. Try again later
Operation(s) failed: [0x8] The server is out of memory. Try again later

bucket is set to full ejection. I am wondering what I am doing wrong.
Current cluster size has 4 nodes. With 8 Cores each, and 14.2 GIG RAM. SSD drives.
CPU/RAM and I/O are fine, but I have the error messages.