23-08-24 01:40 PM
I have observed an issue with Blue Prism using Microsoft edge whereby performance appears to decline. Processing times slow, I suspect due to screen elements taking longer to identify with each repeat of steps.
Has anyone else had issues with Edge not releasing memory when used by Blue Prism Extension or any particular element / issue that degrades performance as below?
I set up a test process that ran a loop of about 5 actions that searched an application, navigated screens, read screens, returned home then closed additional tabs opened by the application. The same reference was searched and same data was read every time. What appears to happen is that with each loop the processing time increases so that an action taking 1min takes 3mins after approximately 15 loops, this means the processing time for all actions increases substantially impacting a business process.
I observed that when the edge browser is exited/closed the time taken resets (returned to initial time to perform) but then starts to build again with each loop of the steps. I also checked task manager and it appears the memory in edge also builds when I am stepping BP actions, which implies increasing memory in Edge could be leading to a reducing processing speed unless I close/open the browser frequently.
I am using a mix of Browser and UIA elements in the actions. I have tried a few things like detaching/attaching, Garbage Collect and a few changes to steps like the order I close the Edge tabs (opened by app) but this has no positive effect - only closing the browser resets the process speed (log out and in app).
I note that I do have an occasional issue with launching edge (BP can freeze, be slow or fail) so have been using the run process utility and attaching to the open browser which resolved launching issues.
We are using Blue Prism 7.3.0.9110 with Browser Extension 7.3 running Edge Version 127.0.2651.86 (Official build) (64-bit)
Answered! Go to Answer.
27-08-24 04:28 PM
To briefly answer your question about whether I've seen that issue, I would say I haven't ever seen it, at least not exactly as you describe. I have seen something like it happen in debug mode after many much looping, but even then it wasn't to the extent that you're seeing it slow down.
Have you reproduced this on multiple machines/networks/environments/websites? What website are you targeting? Is it publicly accessible so I could go try it as well? Have you tried Chrome (if you are able/allowed to) just to see if it has the same issue?
You mentioned you're using a combination of Browser mode and UIA elements. Do you see the same kind of slowness occurring for both element types? If you tested not using any UIA or where it doesn't use any Browser mode elements, do you still see the same problem?
27-08-24 04:28 PM
To briefly answer your question about whether I've seen that issue, I would say I haven't ever seen it, at least not exactly as you describe. I have seen something like it happen in debug mode after many much looping, but even then it wasn't to the extent that you're seeing it slow down.
Have you reproduced this on multiple machines/networks/environments/websites? What website are you targeting? Is it publicly accessible so I could go try it as well? Have you tried Chrome (if you are able/allowed to) just to see if it has the same issue?
You mentioned you're using a combination of Browser mode and UIA elements. Do you see the same kind of slowness occurring for both element types? If you tested not using any UIA or where it doesn't use any Browser mode elements, do you still see the same problem?
28-08-24 10:49 AM
Thanks Dave for taking the time to respond. The website I am using is a new business application being tested, so it may not be publicly accessible (or at least not without user access being granted). I had initially wondered if it is something about the way this target app is built (which can be a little problematic to spy at times) but initially seemed to clears the edge memory better when worked manually.
I've run on a few different VMs in our DEV and Test Environments with the same result but really only really this one process/application. So I agree with your suggestion that I need to see this issue on a different website/browser based app, also consider Chrome to compare and drill down a into those element types a bit further.
I had tried to measure a more simple webpage using browser elements that showed a some hint of slowing by going from 6 secs down to 9 seconds per case over 6hours / 5000 cases in one session but was not really a great comparison alone.
We are only just now in the process moving to BP7 and Edge (from BP5 and IE) while at the same time our business is working to change all its business applications to one new platform, so we don't have a lot of different webpages/apps build out in the new environment yet. I wanted to see if this was a common issue with any obvious edge settings or common issues to target before I started to drill down a bit further.
I noticed that I have not been using the Tracking ID in the edge actions but since I was only using 1 browser and it all appeared to be working well beyond this performance test that did not seem to matter... I guess I can backfill that too to see if makes any difference.
02-10-24 05:09 AM
Hi @Simon.Still,
I am also facing the performance issue using Edge. Have you got any solution to resolve this ?.
02-10-24 03:06 PM
I have seen a degradation in performance before, but not in edge. Where we saw it in the past (many years ago) was with some Java applications and the solution was to periodically (i.e. every 30 minutes or every hour) close down the application fully (making sure the process was closed) and then restarting it afresh. That seemed to solve the issue for us back then and may well work for you today.
Some investigation that might be worth trying is checking out the memory usage of Blue Prism and of your Edge application over time to see if memory is increasing which would suggest a memory leak somewhere (which might not be a Blue Prism issue, it could be a rogue code stage in your solution doing something naughty). Blue Prism has some great memory and CPU logging functionality you can turn on for individual resources to help you investigate.
Also, you didn't mention over what time period this is happening - minutes, hours, days, weeks? Good practice would be to have an overnight reboot of your robot VDIs built in to your standard robot maintenance. In the same way your laptop will slow down and strange things will happen if you never restart it, the exact same is true of your VDIs.