Re: Open socket handles build up over time (leaking?)

2021-11-17 Thread Darrel Schneider
Yes, open a geode bug ticket From: Leon Finker Sent: Wednesday, November 17, 2021 1:13 PM To: dev@geode.apache.org Subject: Re: Open socket handles build up over time (leaking?) Thank you! I'll try to test this change. What's the procedure for this? Should I

Re: Open socket handles build up over time (leaking?)

2021-11-17 Thread Leon Finker
Thank you! I'll try to test this change. What's the procedure for this? Should I open geode bug ticket? On Wed, Nov 17, 2021 at 2:22 PM Darrel Schneider wrote: > > I think you found the leak! > My understanding of the code in registerClientInternal (I'm looking at the > current develop branch)

Re: Open socket handles build up over time (leaking?)

2021-11-17 Thread Leon Finker
Yes definitely not sharing a durable ID between users. Durable ID has login id@host and some other attributes. Even on the same machine if a user tries to run the 2nd instance of the client application, then a unique ID will be generated and appended to the durable ID. It seems the durable

Re: Open socket handles build up over time (leaking?)

2021-11-17 Thread Anthony Baker
I’m curious - how do you assign a durable client id? The id should be unique to a client and not shared among a pool of clients. Anthony > On Nov 17, 2021, at 11:22 AM, Darrel Schneider wrote: > > I think you found the leak! > My understanding of the code in registerClientInternal (I'm

Re: Open socket handles build up over time (leaking?)

2021-11-17 Thread Darrel Schneider
I think you found the leak! My understanding of the code in registerClientInternal (I'm looking at the current develop branch) is that when it logs the warning "Duplicate durable clients are not allowed" that it considers the current client connect attempt to have failed. It writes this

Re: Open socket handles build up over time (leaking?)

2021-11-17 Thread Leon Finker
Following Darrel's excellent advice :) I think I tracked down the area of the socket handle leak. As the article suggested, I ran the lsof capture every 5 minutes. I then traced back the cleaned up socket handles to the valid lsof entries. I verified a bunch of them and they all ended up pointing