Hello, Our service (Truman) is using Go client APIs to read (and write) to Bigtable. The nature of our traffic is such that periodically we will issue a large amount of single row reads from one job to a replicated (2 cell) bigtable in response to user requests.
In my tests i'm seeing very high bigtable read latencies using a single task in the job thats accessing bigtable when a certain QPS is reached. However, increasing the number of tasks, effectively distributing the read traffic for the same total QPS, seems to alleviate this issue. Is there a QPS limit or throttling going on in the client lamprey/envelope on a per task basis? Perhaps some setting i've missed? Dapper traces show a lot of time spent in LampreyBigtable.ForRead. Reference cl/127674083. Most of the relevant stuff is in pods.go, store.go and template.gcl Any thoughts appreciated. Thanks -Rohit -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
