2

On a AWS Master slave hot standby replication schema on Amazon RDS with a single master and a single slave. Both master and replica are being hosted over Amazon RDS on eu-central-1 and the master is used for reading as well. For that we use a rotating dns record that is aliased both on slave and on master database as well.

On my slave database I have the following setting as well:

hot_standby_feedback=on

I receive the following error on the replica:

SQLSTATE[HY000]: General error: 7 SSL SYSCALL error: EOF detected (SQL: ^some_query^)

Also, after further investigations I found out that the error happen both on insert and select queries as well.

And I try to debug this error. The one issue I thought was the connection capacity that may cause memory starvation. But I was proven wrong:

SHOW max_connections;

That has value 1722 and the total active connections are about ~17 (without the one that does replication) on my slave database.

Also in the memory in the slave database has no drop as well it stays in a leveled position as seen on this image:

Memory Level on my Database

Both master ans slave database instances are db.t3.xlarge ones. Hence, it has 4vCPUs, 16GB ram and a 100GB Storage. So being an issue or resources is out of question.

Therefore, I wonder how does the hot_standby_feedback=on may trigger this error and how? Also does the replication may cause unexpected connection termination as well because I use a single replica instead of 2 ones?

In the latter case how I can list past server-terminated connections between a time range?

The database connections are performed over a php Laravel application worker. There are 2 servers that run 4 processes each and for queuing system is used Amazon SQS.

6
  • One cannot be 100% certain, but I'd bet that there is no connection: all that the parameter does concerning communication is that it sends a little extra information back to the master. Can you capture the error and get a stack trace? Commented Feb 6, 2020 at 12:00
  • The error is snown like that and because on Stacktrace has production data as well I cannot display more than this. Commented Feb 7, 2020 at 10:57
  • Ok, but what process gets the error? The WAL receiver? A client backend? Surely you can disclose some of that information. The stack trace would also be valuable if all sensitive string arguments in it were obfuscated. Commented Feb 7, 2020 at 11:07
  • I updated the information regarding the error. I mentioned more info regarding the in stances. Commented Feb 7, 2020 at 11:16
  • Thanks. But where does that error show up? During inserts on the primary? During queries on the standby? During replication? Which exact component receives the error, and what is it doing at the time? Commented Feb 7, 2020 at 11:28

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.