Hi all, I'm trying to understand how many parallel clients will be supported by the queryable state.
so, if I choose 0 for both these parameters, what will be the maximum number of parallel clients supported? I tried more parallel clients than number of slots, but all of them were able to query the state in parallel. Can someone help me to understand the logic here? Thanks in advance. Best Ziyad |
Hi,
I think the number of network treads and number of query threads only roughly correlate with the number of clients that can query in parallel since this is using asynchronous communication via Akka/Netty. Of course, increasing that number means there can be more connections but I think even just 1 Thread or each of those should be able to easily handle multiple queries at the same time. I'm cc'ing Ufuk and Kostas who might know more about this. Best, Aljoscha
|
This is as Aljoscha describes. Each thread can handle many different
clients at the same time. You shouldn't need to change the defaults in most cases. The network threads handle the TCP connections and dispatch query tasks to the query threads which do the actual querying of the state backend. In case of the RocksDB backend for example this might involve blocking I/O calls. – Ufuk On Mon, Aug 14, 2017 at 12:35 PM, Aljoscha Krettek <[hidden email]> wrote: > Hi, > > I think the number of network treads and number of query threads only > roughly correlate with the number of clients that can query in parallel > since this is using asynchronous communication via Akka/Netty. Of course, > increasing that number means there can be more connections but I think even > just 1 Thread or each of those should be able to easily handle multiple > queries at the same time. > > I'm cc'ing Ufuk and Kostas who might know more about this. > > Best, > Aljoscha > > On 9. Aug 2017, at 17:19, Ziyad Muhammed <[hidden email]> wrote: > > Hi all, > > I'm trying to understand how many parallel clients will be supported by the > queryable state. > > query.server.network-threads: number of network (event loop) threads for the > KvStateServer (0 => #slots) > query.server.query-threads: number of asynchronous query threads for the > KvStateServerHandler (0 => #slots). > > so, if I choose 0 for both these parameters, what will be the maximum number > of parallel clients supported? > > I tried more parallel clients than number of slots, but all of them were > able to query the state in parallel. Can someone help me to understand the > logic here? > > Thanks in advance. > > Best > Ziyad > > |
Hi Aljoscha, Ufuk Thank you for the replies. I'm using RocksDB state backend. Could you please explain the blocking I/O calls mentioned? when will it happen? And what will be the effect? A timeout exception? Best Ziyad On Mon, Aug 14, 2017 at 10:17 PM, Ufuk Celebi <[hidden email]> wrote: This is as Aljoscha describes. Each thread can handle many different |
You don't have to worry about this as Flink takes care of all of this.
I was just trying to explain what the query threads are there for. If we would only use the network threads for the queries, these might be blocked during queries and hence not answer network events. But since we use these two different threads pools, we are fine. In the worst case if all query threads are occupied the query will be answered only after a thread becomes available but the network threads are able to accept queries during this time. The summary is that you can probably keep it at the defaults if you don't experience any issues. On Mon, Aug 14, 2017 at 10:44 PM, Ziyad Muhammed <[hidden email]> wrote: > Hi Aljoscha, Ufuk > > Thank you for the replies. > > I'm using RocksDB state backend. Could you please explain the blocking I/O > calls mentioned? > when will it happen? And what will be the effect? A timeout exception? > > > Best > Ziyad > > On Mon, Aug 14, 2017 at 10:17 PM, Ufuk Celebi <[hidden email]> wrote: >> >> This is as Aljoscha describes. Each thread can handle many different >> clients at the same time. You shouldn't need to change the defaults in >> most cases. >> >> The network threads handle the TCP connections and dispatch query >> tasks to the query threads which do the actual querying of the state >> backend. In case of the RocksDB backend for example this might involve >> blocking I/O calls. >> >> – Ufuk >> >> On Mon, Aug 14, 2017 at 12:35 PM, Aljoscha Krettek <[hidden email]> >> wrote: >> > Hi, >> > >> > I think the number of network treads and number of query threads only >> > roughly correlate with the number of clients that can query in parallel >> > since this is using asynchronous communication via Akka/Netty. Of >> > course, >> > increasing that number means there can be more connections but I think >> > even >> > just 1 Thread or each of those should be able to easily handle multiple >> > queries at the same time. >> > >> > I'm cc'ing Ufuk and Kostas who might know more about this. >> > >> > Best, >> > Aljoscha >> > >> > On 9. Aug 2017, at 17:19, Ziyad Muhammed <[hidden email]> wrote: >> > >> > Hi all, >> > >> > I'm trying to understand how many parallel clients will be supported by >> > the >> > queryable state. >> > >> > query.server.network-threads: number of network (event loop) threads for >> > the >> > KvStateServer (0 => #slots) >> > query.server.query-threads: number of asynchronous query threads for the >> > KvStateServerHandler (0 => #slots). >> > >> > so, if I choose 0 for both these parameters, what will be the maximum >> > number >> > of parallel clients supported? >> > >> > I tried more parallel clients than number of slots, but all of them were >> > able to query the state in parallel. Can someone help me to understand >> > the >> > logic here? >> > >> > Thanks in advance. >> > >> > Best >> > Ziyad >> > >> > > > |
Free forum by Nabble | Edit this page |