Will Glass's questions

I have a system which is load balancing via a cookie, but I want to use a URL parameter as an alternative. (Specifically, I have a front end app which does not support cookies). My setup is a little complex - I have a multiplayer system and I use a stick table with this cookie to make sure different people playing the same game go to the same server. This works great, except that the URL parameters don't seem to do anything. This is all with haproxy 1.5dev17

My relevant backend is:

defaults
   log            global
   mode           http
   option         httplog
   option         dontlognull
   retries        3
   option         redispatch
   option         http-server-close

   maxconn        2000
   contimeout     5000
   clitimeout     50000
   srvtimeout     50000

backend simulate
    option httpchk OPTIONS /simulate/api/status
    stick-table type string len 40 size 5M expire 30m
    stick store-response set-cookie(SIMULATE_STICKY_SESSION) table simulate
    stick on cookie(SIMULATE_STICKY_SESSION) table simulate
    stick on url_param(SIMULATE_STICKY_SESSION) table simulate

    server app1 10.0.2.11:8080  cookie app1 check inter 10000
    server app2 10.0.3.11:8080  cookie app2 check inter 10000

Note that SIMULATE_STICKY_SESSION is generated by my app (based on player team).

When I call the URL, I need to pass a jsessionid (so that Tomcat will find the correct session). So I am doing

http://my.domain.com/myapp/url;SIMULATE_STICKY_SESSION=1918C909D3F30DB09C2F96B75C57D015.app1+jsessionid;jsessionid=1918C909D3F30DB09C2F96B75C57D015.app1

This doesn't seem to be picking up the sticky session parameter as 50% of the time it goes to the wrong server. I tried using the param as a query string (after a ? mark) but that didn't work either. What am I doing wrong?

I want to load balance web requests with HAProxy. The catch is that I want particular groups of users to always go to the same server. (This is a multiplayer system, and users in the same group need to be on the same server). The users can go to any server, as long as the users with the same group id are on the same server. The server can even change (for example if a server goes down) but then all users of that group must switch to a new server. The app server will identify the user's grouping with a cookie.

As a specific example, suppose my application has a cookie GROUP_ID numbered 1..10000. There might be 3 different users with GROUP_ID=1, a different 3 users with GROUP_ID=2, etc. And let's assume I set up load balancing on servers: app1, app2, app3.

What I'd like to see is something similar to this:

• all GROUP_ID=1 users go to app1
• all GROUP_ID=2 users go to app2
• all GROUP_ID=3 users go to app3
• all GROUP_ID=4 users go to app1
• all GROUP_ID=5 users go to app2

etc.

If server app1 becomes unresponse, then I want to see the app1 users redistributed among app2 and app3. (It'd be ok if all groups were reshuffled). All users from a particular group should still be on a common server.

Obviously the initial request to the site will not have a cookie, but once the user authenticates they will be assigned the cookie.

Is this kind of custom load balancing supported by HAProxy?

I have a 1 TB MySQL database that I'd like to dump and reload. Most of the data is in one table. A lot of data has been deleted, so I'm pretty sure if I dump it with mysql, rebuild the database, then reload it the total size will be smaller.

I'm dumping the data with this command:

mysqldump  -uroot -pXXX  mydb | gzip -c > data.sql.gz

I get this error

 mysqldump: Error 2013: Lost connection to MySQL server during query when dumping table `MY_TABLE` at row: 596089342

I've tried many variations, including increasing the packet size, doing single translation and going over TCP/IP instead of local socket.

mysqldump  -uroot -pXXX -h 127.0.0.1 --max-allowed-packet=1024M  --single-transaction mydb | gzip -c > data.sql.gz

Finally, I even ran the command going to /dev/null to make sure it wasn't gzip. All permutations produce the same error.

mysqldump  -uroot -pXXX -h 127.0.0.1 mydb > /dev/null

Here's some of the settings in my.cnf

max_allowed_packet = 1G
interactive_timeout = 600
wait_timeout = 600
net_read_timeout=600
net_write_timeout=600
connect_timeout=600

One other odd thing is that the dump always stops on the same place. Approximately 6GB of gzipped data and at approximately the same record. When I do ls -l the file size is always identical.

I'm stumped. Any suggestions for next steps?

For the record, this is Mysql 5.1.58 running on Ubuntu 11.10

WILL

I have a server application which I think is leaking file handles.

I want to track the usage of file descriptors over time on my Linux (ubuntu) server.

I've figured out that I can track the number of file descriptors in use by a process with

lsof -p `pgrep the-process-name` | wc -l

Since I'm already using sysstat and sar to track various metrics, I thought it'd be nice to display with sar. I want to measure this every 10 minutes. Is it possible to add a custom metric to sar? Then I can easily report it out.

If not, I'll write a simple cron job to collect this data and store it separately in a log file.

I want to set Nagios up to issue warnings (e.g. mildly high load or disk usage) via email, but for critical items to alert by pager.

Right now we notify via email and pager simultaneously for all warnings. My generic contact definition is as follows:

define contact{
    name                            generic-contact
    service_notification_options    w,u,c,r,f,s
    host_notification_options       d,u,r,f,s
    service_notification_commands   notify-service-by-email,notify-service-by-pager
    host_notification_commands      notify-host-by-email,notify-host-by-pager
    register                        0

    service_notification_period     24x7
    host_notification_period        24x7
}

How can I make it so that the email notification occurs for both warnings and critical, but paging for critical only?

I'm having problems with sessions accumulating. I run an instance of Tomcat 6.0.18 on a Windows server. I have 3 apps.

I've been having memory issues (close to heap size, or OOM). When I look at the Tomcat manager, I see 500-1000 sessions in each of my 3 apps. When I view the sessions in the Tomcat manager, I can see that some of the sessions are multiple hours in idle time, yet they weren't expired.

One of webapps has a 10 min timeout, but is set in code to 60 minutes (for some users). The other apps have 15 and 30 min timeouts. I also have an issue in that webapp 1 is getting repeated hits with the wrong session cookie (it has to do with my mod_jk load balancing), which drives up the session count. (Lots of sessions created with a single request that then have a 60 min timeout). But it puzzles me that webapp 2 and 3 also are getting high sessions, and that they stick around even past the timeout.

I'm theorizing that the session cleanup is only happening when the server is not heavily loaded, and that (in this case) so many sessions accumulate that the server has low free CPU and the cleanup never runs. But I can't find any references to back this theory up.

Does anyone know the details of the session expiration algorithm? It doesn't run exactly on the timeout minutes -- when does it run?

Possible Duplicate:
What is the correct temperature for a server room?

We're a small company with a 200 foot back room we use as a server room. It has no fans and no ventilation.

Right now, I'm running a small desktop computer as a PBX box and a Dell T605 server for general office use. (We're making heavy use of virtual machines but are basically out of capacity).

I want to add another Dell server to the room. I'm worried though about temperature.

Realistically, at what point do I need to invest in air conditioning for the room? Currently, the temperature is slightly warm when the door is closed but not excessive. How can I effectively measure the need?

I have a table with about 300M rows in MyISAM format I want to convert to Innodb

My original goal was to reduce usage by changing the table schema to have simpler indexes. I dumped all the table, dropped it, recreated it with less indexes, and am now reimporting. However, I forgot to specify that it should be innodb instead of myisam.

Can I just do the standard ALTER TABLE ... ENGINE=INNODB? Is there anything special I should be aware of with such an enormous table?

The data import operation is taking around 12 hours -- I'm loathe to do that again. Hence why I want to just convert it.

I've just set up ESXi 4.0 on my office server. I'm running a couple of vm's from my main 250GB drive.

I want to create a backup of content in the virtual machines on the other 500GB drive. Right now I'm thinking of making a "backup vm" which would hold rsynced copies of key directories from all my other vms.

How can I make a vm that uses this second drive. It looks like I'd need to make a new datastore and put a VM on that datastore -- but I can't figure out how to add a new datastore.

I need to change the the length of a column in a MySQL innodb table with 164M rows. Here's the script I want to run: ALTER TABLE SESSION_DECISION CHANGE COLUMN NAME NAME VARCHAR(255);

When I tried running the alter query, the database churned for an hour and didn't complete the task. Users accessing the system (who needed to write to the table) were locked out. This is a live system so I ended up ctrl-c'ing the request to let users access the site.

(this was in the middle of the night-- usage was minimal but I was getting nervous).

I'm on MySQL 5.0.77.

Any suggestions as to how I can change the column with a minimum of downtime?