From 4ac5dd15521ed48e97be65577348f5ef3f63533b Mon Sep 17 00:00:00 2001 From: David Turner Date: Sun, 16 Mar 2025 20:43:14 +0000 Subject: [PATCH] Mention zero-window state in networking docs (#124969) (#124970) Clarify that it is expected sometimes to see inter-node connections sending zero-window advertisements as part of the usual TCP backpressure mechanism. Backport of #124967 to 8.x --- docs/reference/modules/transport.asciidoc | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/docs/reference/modules/transport.asciidoc b/docs/reference/modules/transport.asciidoc index fc7b6831ca84..b7c0282384e7 100644 --- a/docs/reference/modules/transport.asciidoc +++ b/docs/reference/modules/transport.asciidoc @@ -195,6 +195,15 @@ setting `transport.ping_schedule` if keepalives cannot be configured. Devices which drop connections when they reach a certain age are a common source of problems to {es} clusters, and must not be used. +If an {es} node is temporarily unable to handle network traffic it may stop +reading data from the network and advertise a zero-length TCP window to its +peers so that they pause the transmission of data to the unavailable node. This +is the standard backpressure mechanism built into TCP. When the node becomes +available again, it will resume reading from the network. Configure your +network to permit TCP connections to exist in this paused state without +disruption. Do not impose any limit on the length of time that a connection may +remain in this paused state. + For information about troubleshooting unexpected network disconnections, see <>.