garbagecollect

AusMarkB

Anyone have any recommendations for implementing garbagecollect in this environment?
I've just had our system fall over with
A restart of the socketserver and everything is good again but I think this is the second time in about a fortnight.
Last time there was no messages of value, just out of memory errors. This time there was lots of those too plus this little nugget so that's a little better for troubleshooting.
I think flushing and garbagecollecting might help prevent this moving forward but I don't want to be doing one every request if I don't have to.
Any suggestions?

DonBakke

I seem to be missing something. You start out by asking for "recommendations for implementing garbagecollect" but then end with "I don't want to be doing one every request if I don't have to". This seems like a mixed message.

Before you reply back, please check your copy of HTTP_Services and let me know if you have a ClearSettings service.

AusMarkB

Hey Don,
sorry for the confusion.
Yes I have a clearSettings service.

The mixed message was because I think (but I don't know) that the garbagecollect might be

1. extra overhead I don't want happening too often if I can avoid it and
   
2. may clear caching that is advantageous for the app to have around
   

With that in mind I was just wondering if there was an existing means by which to say flush every nth request or something similar.
Of course my concerns may be overstated and I should just throw a garbagecollect in the mcp routine and be done with it.

Of course, there's also the possibility that it will do absolutely nothing to address the primary issue anyway.

AusMarkB

Now that I think about it, I attach and detach dbs with each request and because they may be different it's probably safest to flush and garbagecollect each request anyway.

DonBakke

If you have that service, then check your existing HTTP_MCP and you should see that this gets called automatically as one of the last couple of things it does before returning the response. You'll see that it performs a Flush/GarbageCollect for you if you have the Flush Cache flag set in your environment. I pretty much have that flag turned on by default. I seriously doubt it will add noticeable overhead.

But, as you say, it might not address the underlying issue.

DonBakke

Now that I think about it, I attach and detach dbs with each request and because they may be different it's probably safest to flush and garbagecollect each request anyway.

Yeah...I was about to make a comment along those lines. You want to keep each request as stateless as possible. To that end, this is why I have the ClearSettings and the CleanUp services being called at the end.

AusMarkB

I see except I must have an early version of clearSettings because there's no mention of flushing cache
//---------------------------------------------------------------------------------------------------------------------- // ClearSettings // // @@DEFINE_SERVICE(ClearSettings) // // Clears all of the global common variables used to track header names, values, and the status settings. This will // typically only be called within HTTP_MCP when the response is finished and sent back to the OECGI. //---------------------------------------------------------------------------------------------------------------------- ClearSettings: Request@ = '' RequestHeaderFields@ = '' RequestHeaderValues@ = '' ResponseHeaderFields@ = '' ResponseHeaderValues@ = '' ResponseStatusCode@ = '' ResponseStatusPhrase@ = '' ResponseBody@ = '' ResponseBodyIsBinary@ = '' SelfURL@ = '' QueryFields@ = '' QueryValues@ = '' Memory_Services('ReleaseHashTable') return
Probably time I should update some things.

AusMarkB

Revisiting this now because after adding the flush/garbagecollect statements it has had no apparent affect on the issue that needs addressing.
So the problem I need solving is how to prevent the engines maxing out the machines memory. The config is set to use ten engines and they all fire up relatively quickly after starting and then stay running as even if there are no direct user requests, there is a task that fires requests every minute so the engines never really become idle enough to shut down.
Everything starts efficiently but over a period of time (days not weeks) the memory usage rises and rises and once it hits about 95% things start to fall over. I need a way to prevent that from happening.
At the moment, I'm simply monitoring and every couple of days as it starts getting closer to the danger point, I restart the socketserver which basically resets everything.
At the moment it's not yet running as a service and granted if it was we could create some automated task to restart the service at regular intervals but I feel that too would just be a bandaid remedy and not a real solution.

I'd prefer it if there was a means of managing the memory without the need to take the service offline even if it is only temporary. I feel like there must be a way, I just don't have any idea what that is.