[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [xmlblaster-devel] Reconnected xml-rpc callbacks result in duplicate messages
Brad Clements wrote:
I've been googling for an answer to this problem, and I'm wondering
what's the best way to handle this.
I'm using Python and pyBlaster.py to subscribe an xml-rpc server for
My xml-rpc server keeps track of pings from the server, and if it
doesn't get any for a while it will re-login and re-subscribe to one key.
Also, I'm running my server under DaemonTools, so if it dies,
daemontools will restart it.
It seems that if my server gets restarted, xml-blaster sees it as a
new login (but with the same callback information). It then sends two
messages where I expect only one.
And, if I restart my server again, then I'll get 3 messages and so on.
I know this is a hard problem for xmlblaster to deal with, especially
since I can subscribe on behalf of another callback listener, so how's
xmlblaster supposed to know what to do?
xmlblaster thinks my server is still logged in (because the client
doesn't get the chance to logout in this particular case), and the
client subscribes to the same key every time.
So, there's two things I need to do:
1. stop multiple duplicate messages
I've been going through the reference guide, and finding a complete
list of QoS parameters is hard.
I know I whined about this last year (sorry), but it sure would be
nice to have a complete QoS list on one page in the docs.
And, having Docs not in a frame would make it easier to bookmark.
Yes, you can work around this by opening the frame in a new tab or window
and then bookmark only the frame.
Michele has designed a new frameless xmlBlaster homepage,
but hadn't time to do the last little issues.
Sooner or later the new homepage will pop up.
if you subscribe multiple times, only one is considered.
Anyway, I found
It works well for EXACT subscribes and XPATH subscribes.
If the later subscribe contains another subscriber-filter
setting, this new filter will be used for future messages
(the subscription is reconfigured)
(see mime based filtering using the message content like for example REGEX).
One drawback for some situations could be that the first
unSubscribe() will remove ALL previous subscribes at once.
Probably we should add a future flag to switch on/off reference counting
so that the same amount of unSubscribe() calls is needed as subsribe() calls
I think you should use multiSubscribe in your case.
if you publish several identical messages on the same topic you only
will get the first update of them (not really what you want here).
No, it is like that:
duplicateUpdates=false will suppress duplicate subscriptions
but it will work only for EXACT oid subscribes.
This setting has a reference counting, you need as many
unSubscribe() calls as the subscribe() calls before.
I would like to set this setting to deprecated and remove
it in a future version as it 'multiSubscribe' has more functionality
and is specific for each subscribe() where as duplicateUpdates is
a general setting for the connection.
Which of these is better to use in my situation? I only ever want one
copy of any single message. I think multiSubscribe=false is the better
choice, but neither option mentions the other one and under what
conditions I'd pick one over the other.
2. Cleanup the old logins to avoid resource exhaustion
don't need to do that (see below)
my server logs in, subscribes and makes no further requests on the
client connection. So, it's sessionTimeout = 0Yes there is:
I login with the same userid every time. Is there any way I can say
"hey, only one login allowed at a time and each new login will logout
the previous instance and clear it's callback". ?
maxSessions=1 in the conntectQos, however I believe you would do better
here to use persistent subscriptions and avoid to re-subscribe, see
and don't forget to use positive sessionId, that way you make sure to
regain the same session and not a new one.
For example login as 'joe/1', that is use always the same
public session id '1' and your above settings and you should always
reconnect to the same server session.
Ah, looks like <session timeout="0" clearSessions="true" /> is what I