Database crashes periodically

[Ed Bailey]

We are having a problem where our cloudSQL database crashes periodically.

The error we are seeing in the logs is: 

[ERROR] InnoDB: Write to file ./ib_logfile1failed at offset 237496832, 1024 bytes should have been written, only 0 were written. Operating system error number 12. Check that your OS and file system support files of this size. Check also that the disk is not full or a disk quota exceeded.

From what I understand, error number 12 means 'Cannot allocate memory'. Is there a way we can configure cloudsql to leave a larger buffer of free memory? The alternative would be to upgrade to have more memory, but from what I understand cloudSQL automatically uses all the memory available to it... Is this likely to reduce the problem or would it likely continue in the same way? 

Are there any other things we can do to reduce this issue?

[Shivam(Google Cloud Support)]

Google Groups discussion forum is meant for open-ended discussions. This sounds like a technical support question which are best suited for Stackoverflow. Make sure to use valid tags. 

Just to mention, it could also turn out to be project/application specific in which case you can consider opting for one of the support packages.

To answer the questions, reducing the buffer(innodb_buffer_pool_size) may resolve memory issues such as this, you can find more details here. About Cloud SQL using all the memory automatically, has to be seen at the project level and depends on the use case that why is it doing so.

[bronze man]

I have see this crash too.It looks like it relate to the mysql server use too much memory.

Use big memory instance will delay this problem, but it can not solve it.

I think it may have too many transactions which is not commit in the memory.

It looks like there is no way to reduce innodb_buffer_pool_size.I can not set it from start, and do not have permission to do that even if I use the root mysql account.

[Jordan (Cloud Platform Support)]

If you are seeing issues with your specific instance, I would highly recommended reporting this in the Public Issue Tracker providing the instance ID and your project ID. Our backend team can then take a look.