https://bz.apache.org/bugzilla/show_bug.cgi?id=61616
--- Comment #64 from Carsten Wolff <[email protected]> --- (In reply to Yann Ylavic from comment #62) > Hmm, I can't reproduce, can you please provide the error log when this > happens? Will do. In the meantime, another observation: Because your patch v12 is constrained to the module code and thus cannot determine, if there still is data in the input filter, it just assumes whenever it cannot write, that there may still be data in the input filter of other. This means, that you call transfer_between_connections every time this happens, while our patch only does that, when there actually is data in the input filters. Since transfer_between_connections does not only drain data from it's input filter but also reads some more data again, we thought you may potentially buffer a lot more data in application memory and allow the sending endpoint to send a lot more, even though you cannot be sure that you will be able to write it again. A simple test with "watch free -m" seems to confirm that, during my test (100 transfers à 10MiB), the application mem with your patch goes up to about 560MiB on my machine, while ours just reaches about 350MiB. -- You are receiving this mail because: You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
