Replace shard localhost RPCs with direct DB access |
||
Issue descriptionEach autoserv run creates a RPC calls for getting label and attribute information and for updating label and attribute information. This creates a lot of RPCs on the shard to localhost because the drone, RPC backend and database are all local. On the shard, especially, it makes a lot of sense to directly call the RPC backend functions to obtain this information from the DB (instead of going through the loopback device and apache). This becomes more viable as we collect the RPC calls discussed here at the beginning and end of autoserv.
,
May 26 2017
Adding motivation for why we want to do this: We've had a number of outages recently where a shard's RPC service became flaky. The fact that a shard depends on its own RPC service makes the shard less reliable. Worse, the shard potentially contributes to its own instability, since many RPC failures are load-related, and the shard is its own most active client.
,
Mar 31 2018
|
||
►
Sign in to add a comment |
||
Comment 1 by pprabhu@chromium.org
, May 26 2017