Date
1 - 2 of 2
Janusgraph Temporary backend operation
|
Misha Brukman <mbru...@...>
Hi Dilan, Looks like you're using the HBase backend, and this is an INFO level exception telling you that when issuing many mutations in parallel ("mutateMany"), some of the mutations succeeded, while some did not. As stated, this is a temporary (retriable) operation, where a backoff strategy typically helps. However, if you are sending a large, steady amount of traffic, there may not be enough time for the cluster to recover from the previous set of operaitons, before the new ones also cause these retriable errors, leading to a cascade of such messages. What this implies is that your application is doing one or both of the following:
If your load is well-distributed, you need to grow your HBase cluster with the size of the traffic you are sending. If you are hotspotting, then growing the cluster may not help, so you may need to use a read-only cache (if read hotspotting), or change your model to avoid sending many concurrent requests to the same set of vertices. Hope this helps, Misha On Tue, Oct 3, 2017 at 12:39 AM, Dilan Ranasinghe <dila...@...> wrote: Hello , |
|
|
|
Dilan Ranasinghe <dila...@...>
Hi Misha, You are right. We are at the requirement of inserting and updating vertices to janusgraph where we need an insert frequency of 500,000 vertices per second and update frequency of 5000 vertices per second. Currently what we could achieve via a poc is 100,000 vertices per second insertions and 450 vertices per second updates using one janusgraph/gremlin server connected to a five node hBase cluster. It is clear from your information that we may need to increase the size of the cluster. Thank you for the information. On Tuesday, October 3, 2017 at 9:42:08 PM UTC+7, Misha Brukman wrote:
|
|
|