I have a Blazor WASM application using .Net Core 6, and all of the latest nuget packages, hosted using Azure App Services.
As per my understanding, the browser caches the application DLLs appending a sha256 hash code to the filename. The browser keeps the DLLs cached until it reads a new sha256 hash value on the blazor.boot.json, at which point it replaces the old DLL in the cache with a new one.
As per my observations, this works most of the time. If I look at my own browser application cache, using the dev tools, I will see the same version the DLL today, tomorrow, and the next day, as long as there hasn't been any deployments. Then, after a deployment, I open the application in my browser, go back to my cache, and without needing to manually clear anything or do anything out of the ordinary, I will see the DLL with an updated sha256 code.
However, every once in a while, this isnt the case. For example, I have a user, that rarely uses the development environment. The version of the application dll has been cached since august (2 months ago). Additionally, the contents of the blazor.boot.json correctly indicate the new version of the dll. But still, the old version, with a different SHA hash, remains in her cache.
I've seen this behavior before. I can't quite put my finger on what the difference is in the application changes that lead to successful automatic cache busting vs when the old cache continues to persist.
The simple solution is to have the user delete the entire blazor-resources group of dll's to force a new download. However, as we expand the footprint of the app to new users, I don't want to have to ask them to clear their application cache as we are deploying new versions of the code.
Given this example, it seems the failure is somewhere between the retrieval of the updated blazor.boot.json and whatever standard javascript runs to retrieve the new DLLs, which I believe is in the service-worker.js.
Any assistance regarding:
An explanation of why this might be happening (so that I can reliably reproduce the issue and have a valid test case after it's been fixed)
What code/configuration I may be able to force the DLL download... not on every refresh of the app, but at least when the SHA hash has changed.
Thanks,
Mike
Related
I have created a webapplication and set up an endpoint which returns a FileStreamResult, these can be large zip and pdf files for example. The main issue I'm facing is that if a download gets interrupted (for example the internet goes down by the client), the temporary file that is generated by the browser is immediately deleted.
I'm aware of range-requests, but to utilize them, I would have to read files from the client, to determine how much progress the download made, which is not possible from the server side and also, the same problem persists about the temporary file deletion.
So this seems like a browser limitation to me, but please correct me if I'm wrong, I would appreaciate any ideas.
At a client of mine, in-house applications are all located on a network share. Users create shortcuts to the required applications from the network share so we can easily make sure everyone uses the latest version.
This works fairly well, although we often have an issue when users are still using applications when we'd like to release a new version. For most applications, we'd forcibly remove all the file locks on the server and release the new version. Not a very elegant solution, especially since we need assistance from another department for this.
For newer applications, I've developed a cleaner solution, where the application intermittently checks if it's still the most recent release. If it isn't, it shows a message to the user, asking him to quit the application at first convenience, or within 3 minutes. After 3 minutes, the application quits itself and all is well. However, some users will immediately try to re-start the application. The application will then show a simple MessageBox telling the user this version is currently not supported. My problem is this: while this MessageBox is visible, my executable is still locked.
I'm looking for any of the following solutions:
Releasing all locks on the current assembly files from within code
Showing a message box that lingers after the current assembly has exited
This is exactly precisely the problem that .NET ClickOnce deployment is meant to solve. Users have a shortcut they can click, the latest version is downloaded on application start, and there are no server-side executables to be locked if a user leaves their process open.
ClickOnce Deployment Overview
HowTo:Publish a ClickOnce Application
We're developing a solution which uses Ektron. As part of our solution we all have local IIS instances (localhost) and deploy to this local instance as part of the development life cycle.
The problem is that after a deployment and once dll's are replaced IIS restarts and the app pool is recycled, this means that Ektron dll's need to reload themselves.
This process takes an extended amount of time.
Is there anyway to improve the loading time of "Ektron"
To some extent, this is the nature of a large app running as a website rather than a web application. Removing the workarea from your local environment is one way to get this compile time down, though this will naturally not work depending on your workflow, for example if you are not using a separate dev DB or if you are storing the workarea in source control.
I have seen some attempts to pre-complile the workarea and keep the working code in a separate project (http://dev.ektron.com/forum.aspx?g=posts&t=10996) but this approach will only speed up your builds, not the recompilation of individual pages that will occur after a build as a result of running as a web site.
The last (and least best-practice) solution is to simply avoid making code changes that cause a recompile, like modifying app_code. Apps running as websites are perfectly happy to recompile a single page's codebehind without regenerating DLLs, which is advantageous for productivity but ultimately discourages good practices like reusing code in libraries. Keep in mind that this is terrible advice, but if you have a deadline and are staring at an ektron page loading every 30 minutes it can be useful to know.
Same problem here. I found this: http://brianpereras.blogspot.com/2013/06/ektron-85-86-workarea-is-slow-compared.html
That says that the help documentation was moved to be retrieved from an online source (documentation.ektron.com). We're running Ektron 9, and I just made this change and it seems much faster on first load (after iisreset).
The solution is to set documentation.ektron.com to 127.0.0.1 in your hosts file.
There is not, this is just how IIS works. Instead of running a local instance of Ektron it's a good idea just to point your web.config file to the database of your test database and copy the /workarea folder to your local PC. You can't edit ektron locally but you can change the data on your test server and it will show up locally.
My web application has been working fine. Just recently, I began trying to add Twitter Bootstrap to my project using NuGet (it doesn't appear that NuGet is the issue because the same thing happens if I add TB manually). After doing so, I noticed that my app was misbehaving ... some items that I was displaying to a page from server-side cache were missing.
As I dug into this, I realized that my app was being re-loaded on almost every call. I placed a break point in BundleConfig and sure enough ... almost every call, I'm hitting the break point.
If I uninstall the package, things start working fine again.
Furthermore, it doesn't seem to be just Twitter Bootstrap. It seems that if I install any new packages into my system, this starts happening ... almost as if I'm pushing IIS Express over some sort of memory boundary?
I've tried to verify some of the normal IISExpress issues with re-compiling ... things being written into bin, etc. But I don't see any activity on that front (and I'm definitely not explicitly writing anything there). I'm not writing to web.config in code or anything either.
Last bit of information -- if I publish the non-working app to my QA server, everything works fine. QA server is full-blown IIS -- not express. This further confirms that nothing is being written into bin or messing with web.config.
EDIT
When I say I added Twitter Bootstrap, I mean that all I did was add it to the project. I haven't even referenced it in any pages. I haven't included it in my bundling/minification, etc. It's basically just sitting there unused but still causes my app to recycle/recompile.
I have created an application that exposes a OSX service for certain file by adding an NSService entry into my applications info.plist (as in http://www.macosxautomation.com/services/learn/), but I find that upon installing my application on a new machine the service doesn't show up quickly in the finder right click context menu.
I know that this is because pasteboard services hasn't re-indexed the /Applications folder and "discovered" the newly installed service.
I also know that I can force a re-index and discovery by manually running /System/Library/CoreServices/pbs.
The question here is what is the best way to ensure that my service shows up as quickly as possible for users who are installing my application for the first time.
I could execute a system call to "/System/Library/CoreServices/pbs" when my application starts up --If the user immediately starts my application--, but that only partly solves the problem (in addition I wonder if there is a better Cocoa API based way of doing this).
If my application is generally only accessed via the context menu, a user will never think to go out and start the application in the first place. They will only think it is broken when the context menu isn't there.
I am not distributing my application with an installer. I am simply providing a bundle that can be dragged and dropped into /Applications (as I believe Apple usually suggests).
Is there a way to expedite the process of service discovery when doing an installation in this fashion, so that there isn't any period of time where the user is without the newly installed service?
As a side note, it appears that the problem may not exist in 10.8 (or at least be as pronounced). Apple may have made this indexing happen more quickly in their most recent release.
I've actually ended up using
system("killall pbs;/System/Library/CoreServices/pbs -flush");
in one of my apps, just as you describe, though it's a long time ago, when 10.5 was in question as well.
You might want to try this function, however:
void NSUpdateDynamicServices(void)
which according to the documentation acts just like flushing pbs, but is a cleaner solution.
Also, if (according to your description), the app is nothing but a service, consider making it a really just a service - see (Installing the Service)
To build a standalone service, use the extension .service and store it in Library/Services.