Rafael H. Schloming commented on PROTON-568:

Messenger will establish a new link for every new address that it sees, and it 
currently doesn't ever attempt to close idle links, so I'm not surprised you 
are seeing this behaviour given the scenario you describe. By the time you get 
to 13000 messages, messenger will be holding 13000 links open.

I'm not sure I would prioritize this super high since I don't think this is 
that common a scenario, but I think there are two possible fixes. One option 
would be for Messenger to have some kind of limit (possibly configurable) for 
how many links it will keep open before closing idle ones. The other option 
would be to try to use the anonymous node so that all the messages could be 
sent over a single link.

> messenger client slows and grows with large address space
> ---------------------------------------------------------
>                 Key: PROTON-568
>                 URL: https://issues.apache.org/jira/browse/PROTON-568
>             Project: Qpid Proton
>          Issue Type: Bug
>          Components: proton-c
>            Reporter: michael goulish
> A messenger-based sending client grows from little memory to over 3 GB, and 
> slows down by at least 10x, when transmitting 1 message each to many 
> addresses.
> here is the setup:
> 1. a single messenger-based sender.
> 2. a single dispatch router
> 3. 5000 messenger-based receivers, each listening to 10 unique addresses.  
> i.e. 50,000 unique addresses total.
> 4. the sender sends a single to each unique address.
> 5. the first 10 messages go to addr1 ... addr10 -- all of which belong to the 
> first receiver.
> 6. when each receiver gets all 10 of its messages, it shuts down.
> 7. each message is 1-to-1, and fire-and-forget.
> 8. This means that the sender only sends 1 message to each address and then 
> moves on, never to return.
> Result
> -----------------------------------
> The sending application started out doing something like 100-200 messages per 
> second.  Its memory footprint was small.  
> By the time I reach 13,000 messages sent, output has fallen to about 25 
> messages per second, and memory (RSS) has risen to about 1.7 GB.
> It keeps getting larger and slower until the user rebels.

This message was sent by Atlassian JIRA

Reply via email to