I've got a change inflight which will allow --afe to be specified to dump_suite_report to switch to a different AFE.  Is an option for TKO required?

Excellent.

Afaik tko overload is not a problem so far, though hitting a readonly replica for that sounds like a prudent idea.

Will shuqianz handle the change to tko/parse or should I?  If me, what's the official name for the slave DB?

Ping on this bug.

So the --afe option will allow us to hit a different rpc server (right? I'm pretty sure your script is making afe rpcs, not direct db queries?)

If so I don't think that will actually help us hit a slave db.

We'd need to make the rpc server smart enough to route read-only requests to the slave db (or, allow the rpc client to pass in a hint about which backend db they want to hit).

+dshi for comment.

Where's the RPC server code?  Is it autotest/files/frontend?

The rpc code is in frontend/afe/rpc_interface.py

We need a new afe role with some special shadow config setting for it to point to slave db.

Are there AFE slave replicas already?

Are there consistency issues with operations that might interleave reads and writes and having those operations hit different database instances?

This change looks like it reverts functionality which previously did this.
https://chromium-review.googlesource.com/223393

dshi@: Can you respond to c#9?

No, secondary AFE role to point to secondary db server is not implemented yet. We need to create a new role for it (most likely just deploying new shadow config file).

The delay between master and slave db should be minimum, like <1min

De-prioritizing since suite report is making use of new index.