I’ve been using Reporting Services 2005 to build a series of BI dashboards, initially using the SharePoint reports library, but more recently using straight ASP.Net pages. In essence we’ve been using SSRS as a headless charting engine.
On some of our dashboards we occasionally saw little red crosses rather than our charts – IE’s way of telling us the image hadn’t loaded. Didn’t happen very often, and at the time we had bigger fish to fry, but unfortunately when we moved to our big new multi-core server, it happened a lot more.
In the logs we saw a lot of this:
w3wp!library!f!01/21/2009-12:40:25:: Call to GetItemTypeAction(http://myserver/reports/BlahBlahReport.rdl).
w3wp!library!f!01/21/2009-12:40:26:: e ERROR: Throwing Microsoft.ReportingServices.Diagnostics.Utilities.StreamNotFoundException: The stream cannot be found. The stream identifier that is provided to an operation cannot be located in the report server database., ;
Hotfix 913363 looked promising, but that was included in SP1, and we were SP3, so unless it had regressed, that wasn’t going to help.
What we determined was that this only happened when the same user requested the same report multiple times (with different parameters) at more or less the same time. And that happens a fair bit for us: our dashboards (both the SharePoint versions, and the later ASP.Net ones) basically rendered down to pages of IFRAMES pointing back at SSRS, and sometimes multiple charts on one dashboard would actually just be the same report re-executed with different parameters. So my ‘race condition’ alarm was off and running.
For any report on the dashboard, IE does this:
- Load the HTML in the IFRAME
Any generated images in the report have to be cached momentarily on the server between when the HTML is generated, and when the browser comes and ask for them. And of course the browser has to supply some kind of reference back to the server so it gets the right images, and not the images from a different report or a what a different user is viewing. So there’s a Reporting Services Session (which you probably read about the first time you got a rsExecutionContextNotFound exception, right?).
So we speculated that SSRS was getting its sessions in a mess, and somehow a subsequent request to the same report, using different parameters, was throwing away the results of the previous execution (the chart image we are after) in the process of creating a new one. At which point Graeme discovered you can disable session cookies (in the ConfigurationInfo table) which (after an IISReset) indeed fixed the problem entirely.
But why the problem in the first place?
Looking at the HTTP traffic between the browser and the server, one sees that the scoping of the image request is indeed only scoped to an imageID and report as part of the GET url, and within a execution session as part of a cookie:
GET /ReportServer?BlahBlahReport&rs%3aFormat=HTML4.0&rs%3aImageID=d95ebcc7-deba-4734-93c9-270468bd133b HTTP/1.1
Cookie: ConsoleVisible5dd14218-294b-424e-a33a-013236be5290=false; RSExecutionSession%3ahttp%3a%2f%2fserver%2f%2fBlahBlahReport=54m0qr55hytlvn55nzkjxr55
I’ve not followed this through entirely, but basically the trouble is your browser only has one RSExecutionSession cookie, which appears to contains only one execution ID per report URI. You’d imagine then that what you’d get is just the wrong image (outstanding requests for images from the first execution are submitted using the execution ID from the second session), but you don’t, so the most likely scenario I can come up with is that starting the second execution for the report implicitly removes the artefacts (images etc…) still in cache from the first execution. If this happens before the browser’s loaded them then it’s little broken crosses time…
Once you turn cookieless sessions on, however, the GET requests look more like this:
GET /ReportServer?%2fBlahBlahReport&rs%3aSessionID=ucqtvo55wacqugqjln3l2045&rs%3aFormat=HTML4.0&rs%3aImageID=e9a2d08b-23cb-4586-8399-eb42ed5558a7 HTTP/1.1
ie they contain all three of report, execution and image. As such executing the same report multiple times in parallel doesn’t create any race conditions. This seems like such a better way of doing it that I’m surprised this isn’t the default behaviour (I guess it can’t clean up the temporary cache quite as aggressively, but other than that I can’t see the downside).
It’s worth noting that this kind of thing is an absolutely classic pitfall with any form of HTTP session state, which application frameworks like ASP and ASP.Net typically sidestep by serializing requests within a session and thus avoiding order-of-execution issues. However in this case, even if SSRS serialized access it would have to ensure it serialized it in the right sequence to really fix the problem. Tricky.
PS: Reporting Services 2005. Haven’t tested on 2008.