It feels like the wrong usecase for kafka. Its not meant as something you
connect your end users to. Maybe MQTT would be a better fit as the serving
layer to end users or just poll as you said.

2017-07-31 17:10 GMT+02:00 Thakrar, Jayesh <jthak...@conversantmedia.com>:

> You may want to look at the Kafka REST API instead of having so many
> direct client connections.
>
> https://github.com/confluentinc/kafka-rest
>
>
>
> On 7/31/17, 1:29 AM, "Dr. Sven Abels" <ab...@ascora.de> wrote:
>
>     Hi guys,
>
>     does anyone have an idea about the possible limits of concurrent users?
>
>     -----Ursprüngliche Nachricht-----
>     Von: Dr. Sven Abels [mailto:ab...@ascora.de]
>     Gesendet: Freitag, 28. Juli 2017 12:11
>     An: users@kafka.apache.org
>     Betreff: Limit of simultaneous consumers/clients?
>
>     Hello,
>
>
>
>     we would like to use Kafka as a way to inform users about events of
> certain
>     topics. For this purpose, we want to develop Windows and Mac clients
> which
>     users would install on their desktop PCs.
>
>
>
>     We got a broad number of users, so it's likely that there will be
> >10.000
>     clients running in parallel.
>
>
>
>     If I understand it correctly, then Kafka uses Sockets and the user
> clients
>     would maintain an active connection to Kafka. If this is correct, I
>     wondered:
>
>
>
>     -What is the limit of clients that may run in parallel? Do 10.000
> clients
>     mean 10.000 server connections? Would that be a problem for a typical
>     server?
>
>
>
>     -Can we solve this problem by simply running kafka on several servers
> and
>     using something like a round-robin for the DNS so that the clients
> connect
>     to different servers?
>
>
>
>     -We expect to only send a few messages each day. Messages should arrive
>     quickly (<30 seconds delay) but we don't need realtime. Considering
> this: Is
>     kafka still a good solution or should we better switch to e.g. polling
> of
>     clients to the server without Kafka?
>
>
>
>
>
>
>
>     Best regards,
>
>
>
>     Sven
>
>
>
>
>
>

Reply via email to