We run OI 3.7.1 in an O'Reilly webserver context on NT 4 (Service Pack 5).

Every couple of days we get the message that OICGI has caused an application error, typically at 0xc000005

Sometimes we have to reboot the NT. Sometimes the message does not inhibit continued operation of the service via OICGI. We have not been able to determione the cuase. Nothing on the logs.

We will be interested to see what 3.7.3 brings

Have you been scanning for GFEs from the DB Manager in OI? Is there any Revelation Network product is use, here?

FYI: This morning (11.00 am) we completed upgrading from 3.7.1 via 3.7.2 to 3.7.3 (pre-release) to take advantage of the new OICGI. The NT is running OI as a dedicated service

At 5.00 pm, after functioning satisfactorily with accesses every 10 secs, we got an OINSIGHT GPF in Oinsight.exe at 0003:46C9. After OK and reboot of OI, the Gateway went beserk processing a sequence of stacked (?) queries and then the NT froze.

Anthony,

You were pounding the server pretty hard … in fact continuously. The OpenInsight internet gateway window has a timer event that checks to see if the list of requests that it displays is getting to big and if so it clears the list. Since the server was getting hit non-stop, and the timer events in Windows are low-priority, the timer event never fired! As a result, the list over-flowed.

The other behavior you saw is a result of trying to restart after the GPF. We do know how to fix the problem you saw though, and will do so as part of the next pre-release.

So, were you testing the OICGI update? Is that why it was getting hit non-stop?

Cameron Purdy

Revelation Software

Thanks Cameron. In fact we were testing a link checker from static pages on our ISP that have many links pointing into the OICGI as queries. We were checking to see what sort of resistance the site had to being hoovered! To try to find where to put blocks against abusive access.

I am not quite clear what you said you might fix. On our side, I guess we have to prevent excessive efforts at access from a given user. It was a bit of a coincidence, but yes we were running against the update OICGI, which otherwise seemed stable.

Anthony,

The problem you saw is not new in the 3.7.3p1 … it was originally isolated by Byron Schindler. The work-around is to implement INET_SECURITY to check the size of the list of previous requests in the INET_GATEWAY window and (if the list is ]20K for example) to clear the list.

Again, to be clear, this is not an OICGI bug, but rather a bug in the OpenInsight Internet Gateway window that we can fix relatively easily for 3.7.3. The second half of what you saw (after you restarted OI) appears to be a problem in the WOW layer of NT cleaning up the DDE status after a GPF … we did check the code to make sure that there was nothing we were doing wrong (like IPC loops without timeouts) and we could not find any place where OICGI would lock up if the DDE subsystem were working correctly. As a result, I suggest you either reboot after a GPF in OI or OE or OICGI (which hopefully will not happen after we provide the fix) or at least make sure that all the OICGI programs have terminated. (They should time out after 30 seconds.)

Since p1, we have made one or two minor enhancements to OICGI, and I assume those will be provided in an upcoming p2. There are one or two other items that need to be done as well before we finalize these fixes because we want to provide source code too.

I hope that helps to clarify what you saw.

Cameron Purdy

Revelation Software

View this thread on the Works forum...