René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Queued buffers not cleaned up in closed sockets","articleBody":"I initially reported this issue on [ws repo](https://github.com/websockets/ws/issues/667), however there might be a possibility that it is a node issue after testing it on both the official windows Nodejs and [Chakra](https://github.com/Microsoft/node/releases/tag/node-chakracore-6.0.0-pre2) builds. Somehow it seems that the issue can be reproduced more quickly on Linux/MacOS than on Windows.\n### Observations\n- Slow clients (clients receiving data slower than server sends it) causes socket to increase their bufferSize\n- This typically increases the RSS of the node process\n- Upon disconnections of clients, we expect RSS to fall\n- This is true for TCP test case but not for WS (Websockets)\n- Slower RSS recovery for TCP test can be triggered by disconnecting most but not all of the connected clients\n- Recovery of RSS can be sometimes triggered by connecting new clients and disconnecting them quickly\n### Reproduction Steps\n- Test cases are in https://github.com/zz85/node-ws-slow-client-buffer-leak repository\n- Run server on a separate machine over the network\n- Run 200 clients for 30 seconds\n- RSS builds up to approximately 700MB.\n- Disconnects clients.\n- RSS falls to about 300MB and stays there (RSS falls to 60MB for windows nodejs)\n","author":{"url":"https://github.com/zz85","@type":"Person","name":"zz85"},"datePublished":"2016-01-20T11:28:02.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":13},"url":"https://github.com/4779/node/issues/4779"}