I'd like to add one more consideration to Rob's issue. It has been found far to often that a developer "forgets" to provide recovery logic to their applications. I would add the phrase "imperative that programmers make their applications self-healing" to the second paragraph. The recovery logic being addressed in this thread is not rocket science.
Without application recovery it is usually left to some poor operations guy at 3:00am to figure out what went wrong. A little time of thought and effort up front could save loads of time at the back end. Additionally concerning clients.... For applications connecting locally I have seen where QMgr and Queue handles have been corrupted. The same recovery logic used in the clients would be used in server based connections. Sorry for the lateness of this thread... I've been out of the office for a week. Brian Bumpass Wachovia Bank E-Channels [EMAIL PROTECTED] Phone - (704) 590-5620 -----Original Message----- From: Miller, Dennis [mailto:[EMAIL PROTECTED]] Sent: Tuesday, November 12, 2002 5:04 PM To: [EMAIL PROTECTED] Subject: Re: Clients & Network "Glitches" Rob, In progressing through this thread, I became increasingly nervous that either I was out of touch with reality or that what I perceived to be some rather dicey considerations were being overlooked. I was relieved see your thorough and thoughtful response which saved me lots of soul-searching. Thanks. Anyway, I would like to add a couple of points. While it is certainly possible, and possibly advisable, for programmers to make their applications "self-healing" with respect to network glitches, it is not usually such an easy undertaking and there are always practical limits to how much can be done. Not only are the appropriate recovery steps different by application, they are very-well likely to vary across different states of the same application. We end up in a drill that quickly approaches the point of diminishing returns. Furthermore, no matter how much we invest in that exercise, one must always have contingencies for the utlimate possibilty that the network went bye-bye and will not be coming back in an acceptable timeframe. One of the primary benefits of MQ is to make programming easier; case in point--to relieve programmers of the neccessity of handling presumably infrequent, yet critical and unforgiving problems like network glitches. The firm goal should be to address such transient failures in the infrastructure. This is especially true in the context of network failures which inevitably surface as the weakest link. Project the notion of "network" to the Internet--read that: global reach and the responsibility of no one in particular--to really appreciate this point. This leads me to mention the underlying means by which MQ delivers a significant degree of isolation from network anomolies. Of course, I'm referring to asynchronous messaging. The relevant benefit of that technology is that queueing occurs local to the application, thereby reducing the dependence on network availability. The MQ transport gives us assurances that when the network is restored, in progress message flows will complete. Designers can therefore concentrate more on business problems, like removing time dependencies from their applications, blissfully letting somebody else worry about recovery from network failures. MQ Client, compromises this strategy. I don't assert that one should never use it--just understand that it doesn't provide the full complement of MQ benefits. If the requirements are to function during periods of network downtime or to transparently survive network glitches, then I would argue that the client does not serve those requirements very well. Perhaps one should consider using a technology with a better fit, rather than trying to compensate. regards, Dennis > -----Original Message----- > From: Wyatt, T. Rob [SMTP:[EMAIL PROTECTED]] > Sent: Friday, November 08, 2002 2:33 PM > To: [EMAIL PROTECTED] > Subject: Re: Clients & Network "Glitches" > > Jim, > > When you get the 2009 depends on what your application is doing. If you > do > a GET with an extended wait, you may not get the 2009 until the channel > heartbeat interval cycle. If the application does PUTs in response to > user > input, you may not get the 2009 until the PUT happens. It is only issued > if > your application is exercising the MQ API and the API discovers that the > connectivity has been lost. > > Furthermore, the nature of the client is that the client MCA is really a > process at the server that acts as a proxy for your client. The 2009 is a > client response that indicates it can't reach the proxy but doesn't > indicate > what the MCA did. If the last action was a PUT or a GET, it may have > completed at the server - or not - you really don't know. For this > reason, > all client code is advised strongly to do everything under syncpoint. > > Even if you do everything under syncpoint, the same consideration applies > - > if you get a 2009 you don't know the outcome at the server. For PUT and > GET > this is not a problem because the transaction will be rolled back > (eventually). If however you get a 2009 on the COMMIT, you really don't > know whether the transaction completed or not. > > Finally, if you do leave an uncommitted transaction running when your > channel drops, you want to make sure that it eventually rolls back. Be > sure > to use heartbeat in the channel, and the KeepAlive and AdoptNewMCA in the > QM.ini. Without proper tuning, your zombie client MCAs can hang around > forever and hold transactions open. This can cause problems with log > files > overflowing eventually as well as exceeding your max channels. > > As far as your network guys contention that "MQ should be more tolerant of > these problems", they are half right. MQ does a great job of tolerating > these problems if tuned correctly - and by MQ I mean the components IBM > supplies like the MCAs. This is a case where it is the application that > is > not tolerating the outages but the good news is that it sounds like it is > a > homegrown application. Getting vendors to fix their apps can be a > nightmare. > > As someone else said in a previous post, the application should detect and > handle 2009 errors - and possibly others. Some errors you can't do much > about and the app should probably report them and die (i.e. - 2012 > ENVIRONMENT_ERROR). Others like 2016 GET_INHIBITED may be transient in > your > environment (if you regularly GET-inhibit queues to control application > flow > for example) or they may indicate a permanent problem. Once you decide > which error codes merit a disconnect and reconnect you can implement code > to > survive the outage which will make you, your users and your network people > much happier. > > -- T.Rob > > -----Original Message----- > From: Jim Ford [mailto:[EMAIL PROTECTED]] > Sent: Friday, November 08, 2002 10:38 AM > To: [EMAIL PROTECTED] > Subject: Re: Clients & Network "Glitches" > > > There's no doubt that the network people are attempting an end-around. > I just want to be prepared for it. Just to clarify: you get the TCP/IP > "connect reset by peer" error as soon as the network burps. And this > can't be tuned? There's no setting, either in Windows or in MQ, that > can influence this? > > Also, besides reconnecting, the app would also have to reopen the > queues, too, correct? > > > > > Michael F > Murphy/AZ/US/MQSo To: > [EMAIL PROTECTED] > lutions cc: > <mmurphy@MQSOLUTI Subject: Re: Clients & > Network "Glitches" > ONS.COM> > Sent by: MQSeries > List > <MQSERIES@AKH-WIE > N.AC.AT> > > > 11/08/2002 01:05 > AM > Please respond to > MQSeries List > > > > > > > Instructions for managing your mailing list subscription are provided in the Listserv General Users Guide available at http://www.lsoft.com Archive: http://vm.akh-wien.ac.at/MQSeries.archive Instructions for managing your mailing list subscription are provided in the Listserv General Users Guide available at http://www.lsoft.com Archive: http://vm.akh-wien.ac.at/MQSeries.archive
