diff options
author | Brian Cameron <bcameron@src.gnome.org> | 2006-02-09 20:04:43 +0000 |
---|---|---|
committer | Brian Cameron <bcameron@src.gnome.org> | 2006-02-09 20:04:43 +0000 |
commit | 9dbb5bfdb6d063dc98a22403b3095e930dec6ad0 (patch) | |
tree | 644d994e5775f8d3f28d9616cc0cbf001488a8b8 /config | |
parent | 441eb0bd94c240cce85783ffd30b53e6194d4f97 (diff) | |
download | gdm-9dbb5bfdb6d063dc98a22403b3095e930dec6ad0.tar.gz |
A known issue with gdmdynamic is that when a display connects to the
A known issue with gdmdynamic is that when a display connects to
the server it generates the gdmdynamic "ADD" and "RELEASE" commands.
On startup, hundreds of displays could send these commands at once
and cause the server too be flooded with sockets requests. Hammering
the socket like this caused me to find and fix a number of problems
that improve socket reliability for general GDM use. I also enhanced
gdmdynamic so it is more sensitive to the socket being busy and no
longer overloads it, instead sleeping and retrying if necessary. This
allows gdmdynamic to work if there are hundreds of displays instead of
just a dozen or so.
* daemon/gdm.[ch]: Added new "SERVER_BUSY" sockets command so that
gdmdynamic can sleep before starting new slaves if the daemon is
already busy.
* daemon/gdm-net.[ch]: Bump up MAX_CONNECTIONS from 10 to 15. I notice
that this improves performance significantly when hammering the
daemon with connections. Added better comments for this logic and
now debug logs when a subconnection is thrown away. New
gdm_connection_is_server_busy function
* daemon/display.c: Correct DYNAMIC_REMOVE so it works and fixes bug
#326796. Before it wasn't really removing the displays.
* gui/gdmdynamic.c: Quite a bit of work to make gdmdynamic avoid
flooding the server with sockets requests. Now it sets sockets
retries to 1 and manages sleeping and retries itself.
* gui/gdmconfig.c: Added gdm_config_set_comm_retries so that slaves can
specify how many retries they want the comm logic to use.
* gui/gdmcomm.c: Now do_command returns NULL when it gets back "",
which happens when a subconnection was dropped by the daemon. This
lets the slave try the connection again. Now error messages are
always logged, not just when debug is turned on. Added
gdmcomm_did_connection_fail and gdmcomm_set_allow_sleep so
gdmdynamic can control the behavior of how the connection works.
* gui/gdmcommon.c, gui/gdmchooser.c, gui/gdmlogin.c, gui/greeter/greeter.c:
Fix gdm_common_fail so it doesn't generate compile errors when building
with GCC. Fixes bug #330480.
* docs/C/gdm.xml: Cleaned up section that explains sockets commands so
that they are in alphabetical order, added info about SERVER_BUSY
and new gdmdynamic -s and -t options.
* config/gdm.conf: Better description of how debug works, perhaps
I just didn't like the word "spew".
---------------------------------------------------------------------
Diffstat (limited to 'config')
-rw-r--r-- | config/gdm.conf.in | 7 |
1 files changed, 4 insertions, 3 deletions
diff --git a/config/gdm.conf.in b/config/gdm.conf.in index 12ba43e0..c15cb86c 100644 --- a/config/gdm.conf.in +++ b/config/gdm.conf.in @@ -498,9 +498,10 @@ Multicast=false #AllowAdd=true [debug] -# This will enable debugging into the syslog, usually not necessary and it -# creates a LOT of spew of random stuff to the syslog. However it can be -# useful in determining when something is going very wrong. +# This will cause GDM to send debugging information to the system log, which +# will create a LOT of output. It is not recommended to turn this on for +# normal use, but it can be useful to determine the cause when GDM is not +# working properly. Enable=false # This will enable debug messages for accessibilty gesture listeners into the # syslog. This includes output about key events, mouse button events, and |