sinping's questions

It appears as though Firefox requests web fonts (specifically Google's) in a different manner than other browsers. They fail to load when behind Squid as a result, and I'm trying to isolate why.

Using this page as a reference in various browsers, Firefox noticeably displays it differently.

In the Squid logs, you see hits like this:

TCP_DENIED/407 3292 GET http://themes.googleusercontent.com/font? - NONE/- text/html

If you load the same page with IE, you see hits like this:

TCP_DENIED/407 4389 GET http://themes.googleusercontent.com/static/fonts/federant/v1/C109bUmZeyhh-vIXq9lNfvesZW2xOQ-xsNqO47m55DA.woff - NONE/- text/html

So why the difference and how can we make it play nice with Firefox? This is running Squid 3.1.15, but also tested with a few other 3.1.x versions.

I'm curious to hear strategies and methods people use for performing alters on very large tables in MySQL. Large could be any number of rows or size that would be impacting to alter. For the sake of conversation, let's say 2 million+ rows to which any alter is going to be impacting to normal application performance.

The two primary strategies I see are to either perform the alter on a slave and then promote it to be the master after it has finished, or to create a copy of the table with the intended alters already done then copy and catch up the data and do a rename to swap them out before dropping the old.

I am ideally seeking a means to do the latter. A large concern of mine here is triggers on the table being altered, and of course ensuring the data in the two tables is kept in sync before they are swapped. I am thinking the potential for error or missing data can be mitigated to some extent by making use of the read_only variable at key points in the process to make sure data is not changing when fiddling with triggers and after you have caught up all the data. I understand that would have an impact on the application using the database, but it's better than risking corrupted data.

I have been looking at utilities and strategies for doing this and there are several out there. A notable one is this one, which Facebook used as the basis for their online alters. : openark kit documentation. The process is then elaborated upon here: Thoughts and ideas for Online Schema Change

What are your experiences with either method? What pitfalls and gotcha's did you come across? Which do you prefer/suggest and why?

Percona/Maatkit also have their own: pt-online-schema-change

I'm working on setting up a cluster of servers with an iSCSI MD3200i SAN for shared storage. Everything is working well but I have one small detail I can't seem to get working. Multipath seems to only want to do failover with the iSCSI connections to the SAN. I'd like to get this working in load balancing mode so that it uses each path and not just one or the other.

One always shows as ghost here, meaning it's not being used.

[root@kvm-01]~# multipath -ll
mpath2 (36842b2b0006b9d87000004383bf558d9) dm-5 DELL,MD32xxi
[size=2.2T][features=3 queue_if_no_path pg_init_retries 50][hwhandler=1 rdac][rw]
\_ round-robin 0 [prio=100][active]
 \_ 8:0:0:0   sdb 8:16  [active][ready]
  \_ 7:0:0:0  sdc 8:32  [active][ghost]

My multipathd conf:

[root@kvm-01]~# egrep -v '(#|^$)' /etc/multipath.conf
blacklist {
        device {
                vendor  "*"
                product "Universal Xport"
        }
        device {
                vendor  "*"
                product "MD3000"
        }
        device {
                vendor  "*"
                product "MD3000i"
        }
        device {
                vendor  "*"
                product "Virtual Disk"
        }
devnode "^(ram|raw|loop|fd|md|dm-|sr|scd|st)[0-9]*"
devnode "^hd[a-z][[0-9]*]"
devnode "^cciss!c[0-9]d[0-9]*[p[0-9]*]"
devnode "^sda$"
}
defaults {
        user_friendly_names   yes
        polling_interval      5
        selector              "round-robin 0"
        path_checker          rdac
        path_grouping_policy  multibus
        rr_weight             uniform
        no_path_retry         30
        failback              immediate
        rr_min_io             100
        prio_callout          "/sbin/mpath_prio_rdac /dev/%n"
        max_fds               8192
}
devices {
        device {
                vendor                "DELL"
                product               "MD32xxi"
                hardware_handler      "1 rdac"
                features              "2 pg_init_retries 50"
        }
        device {
                vendor                "DELL"
                product               "MD32xx"
                hardware_handler      "1 rdac"
                features              "2 pg_init_retries 50"
        }
        device {
                vendor                "DELL"
                product               "MD36xxi"
                hardware_handler      "1 rdac"
                features              "2 pg_init_retries 50"
        }
}

I've tried a variety of group_by and rr_weight settings, all with the same result.

[root@kvm-01]~# lsmod | grep rdac
dm_rdac                41673  1
dm_multipath           58457  3 dm_round_robin,dm_rdac
scsi_mod              199001  14 dm_rdac,be2iscsi,ib_iser,iscsi_tcp,bnx2i,cxgb3i,libiscsi2,scsi_transport_iscsi2,scsi_dh,sr_mod,sg,libata,megaraid_sas,sd_mod

I've also tried loading scsi_dh_rdac by that didn't make a difference either.

[root@kvm-01]~# egrep -v '(#|^$)' /etc/iscsi/iscsid.conf
node.startup = automatic
node.session.timeo.replacement_timeout = 30
node.conn[0].timeo.login_timeout = 15
node.conn[0].timeo.logout_timeout = 15
node.conn[0].timeo.noop_out_interval = 5
node.conn[0].timeo.noop_out_timeout = 15
node.session.err_timeo.abort_timeout = 15
node.session.err_timeo.lu_reset_timeout = 20
node.session.initial_login_retry_max = 8
node.session.cmds_max = 128
node.session.queue_depth = 32
node.session.iscsi.InitialR2T = No
node.session.iscsi.ImmediateData = Yes
node.session.iscsi.FirstBurstLength = 262144
node.session.iscsi.MaxBurstLength = 16776192
node.conn[0].iscsi.MaxRecvDataSegmentLength = 262144
discovery.sendtargets.iscsi.MaxRecvDataSegmentLength = 32768
node.conn[0].iscsi.HeaderDigest = None
node.session.iscsi.FastAbort = No
node.session.xmit_thread_priority = -20
node.conn[0].iscsi.MaxXmitDataSegmentLength = 0

I've been researching this for awhile now, and I've found plenty of people getting this setup to work with a MD3000i, but no confirmation either way of the 3200i. I found one person saying it doesn't support it because the secondary controller is passive by design but I've been unable to confirm that in Dell's documentation.

[root@kvm-01]~# uname -a
Linux kvm-01 2.6.18-238.9.1.el5 #1 SMP Tue Apr 12 18:10:13 EDT 2011 x86_64 x86_64 x86_64 GNU/Linux