[#CORE-2107] Failed to establish TCP\IP connection with Windows Classic Server under high load

Firebird Core

Failed to establish TCP\IP connection with Windows Classic Server under high load

Created: 04/Oct/08 09:14 AM Updated: 11/Sep/10 09:34 AM

Component/s:

Engine

Affects Version/s:

2.0.0, 1.5.4, 2.0.1, 2.0.2, 2.0.3, 1.5.5, 2.5 Initial, 2.1.0, 2.0.4, 2.5 Alpha 1, 2.1.1

Fix Version/s:

2.5 Beta 1, 2.1.4

Time Tracking:

Not Specified

Environment:

Windows, CS, TCP\IP

Planning Status:

Unspecified

Description

« Hide

When many simultaneous attempts of connection is performed some of them may fail with error 335544421 (connection rejected by remote interface)

Reproduced on Windows Classic Server using TCP\IP

All

Comments

Work Log

Change History

Version Control

Subversion Commits

Sort Order:

[ Permalink | « Hide ]

Vlad Khorsun added a comment - 04/Oct/08 09:17 AM

It seems that if listener process is busy (creating new work process) when few connection attempt is already in some internal Windows queue than Windows may drop some connection attempts.

[ Permalink | « Hide ]

Vlad Khorsun added a comment - 04/Oct/08 02:11 PM

All fork (CreateProcess) calls moved to additional dedicated thread. This allows listener to accept new connection almost immediately after previous. This can't give full guarantee that under very high load all connections would be accepted but improved things a lot. Currently i see no failures with up to 300 simultaneous connection attempts and less than 10 failures with 500 attempts. Before this fix even with 50 cincurrent connection attempts i seen 10-30 failures.

[ Permalink | « Hide ]

Philippe Makowski added a comment - 16/Dec/08 08:24 AM - edited

will it be backported to 2.1.x ?

[ Permalink | « Hide ]

Vlad Khorsun added a comment - 16/Dec/08 10:23 AM

It is not a regression therefore i didn't commit it to the 2.1.2...
If there will be decision to backport it - i would have no problem with it ;)

[ Permalink | « Hide ]

Dmitry Yemanov added a comment - 16/Dec/08 10:30 AM

My suggestion is to release v2.1.2 "as is" and revisit this ticket for v2.1.3 if anybody would report us about being affected by this issue.