Mirror: Difference between revisions

From CSCWiki
Jump to navigation Jump to search
No edit summary
(5 intermediate revisions by the same user not shown)
Line 17: Line 17:
==== Storage ====
==== Storage ====


All of our projects are stored on one of two zfs zpools. There are 8 drives per array (7 run cscmirror3), configured as raidz2, and there is an additional drive that can be swapped in (in the event of a disk failure).
All of our projects are stored on one of three zfs zpools. There are 8 drives per array (7 run cscmirror3), configured as raidz2, and there is an additional drive that can be swapped in (in the event of a disk failure).


* <code>/mirror/root/.cscmirror1</code>
* <code>/mirror/root/.cscmirror1</code>
Line 75: Line 75:
As of Winter 2010, it is now generated by a Python script in <code>~mirror/mirror-index</code>.
As of Winter 2010, it is now generated by a Python script in <code>~mirror/mirror-index</code>.


<code>~mirror/mirror-index/make-index</code> is scheduled in <code>/etc/cron.d/csc-mirror</code> to be run at 5:40am on the 14th and 28th of each month. The script can be run manually when needed (for example, when the archive list is updated) by running:
<code>~mirror/mirror-index/make-index</code> is scheduled in <code>/etc/cron.d/csc-mirror</code> to be run hourly. The script can be run manually when needed (for example, when the archive list is updated) by running:


<code>sudo -u mirror /home/mirror/mirror-index/make-index.py</code>
<code>sudo -u mirror /home/mirror/mirror-index/make-index.py</code>


The script will iterate all folders in <code>/mirror/root</code>, identify the size of the project using `zfs get -H -o value used $dataset`, where $dataset is calculated from the symlink in <code>/mirror/root</code>. The size of all folders is added together to calculate the total folder size (the total size includes hidden projects).
This causes an instance of <code>du</code> which computes the size of each directory. This list is then sorted alphabetically by directory name and returned to the Python script. If any errors occur during this process, the script conservatively chooses to exit rather than risk generating an index file that is incorrect.


<code>make-index.py</code> is configured by means of a [https://yaml.org YAML] file, <code>config.yaml</code>, in the same directory. Its format is as follows:
<code>make-index.py</code> is configured by means of a [https://yaml.org YAML] file, <code>config.yaml</code>, in the same directory. Its format is as follows:


<pre class="yaml">docroot: /mirror/root
<pre class="yaml">docroot: /mirror/root
duflags: --human-readable --max-depth=1
output: /mirror/root/index.html
output: /mirror/root/index.html


Line 103: Line 102:


# (...)</pre>
# (...)</pre>
The docroot is the directory which is to be scanned; this will probably always be the mirror root from which Apache serves. <code>duflags</code> specifies the flags to be passed to <code>du</code>. This is here so that it's easy to find and alter. For instance, we could change <code>--human-readable</code> to <code>--si</code> if we ever decided that, like hard disk manufacturers, we want sizes to appear larger than they are. <code>output</code> defines the file to which the generated index will be written.
The docroot is the directory which is to be scanned; this will probably always be the mirror root from which Apache serves. This is here so that it's easy to find and alter. For instance, we could change <code>--human-readable</code> to <code>--si</code> if we ever decided that, like hard disk manufacturers, we want sizes to appear larger than they are. <code>output</code> defines the file to which the generated index will be written.


<code>exclude</code> specifies the list of directories which will not be included in the generated index page (since, by default, all folders are included in the generated index page).
<code>exclude</code> specifies the list of directories which will not be included in the generated index page (since, by default, all folders are included in the generated index page).


Finally, <code>directories</code> specifies the list of directories to be listed. The format is fairly straightforward: simply name the directory and provide a site (the display name in the &quot;Project Site&quot; column) and URL. One caveat here is that YAML does not allow tabs for whitespace. Indent with two spaces to remain consistent with the existing file format, please. Also note that the directory name is case-sensitive, as is always the case on Unix.
Finally, <code>directories</code> specifies the information of directories. All directories are listed by default, whether or not they appear in this list - only those under <code>exclude</code> are ignored. The format is fairly straightforward: simply name the directory and provide a site (the display name in the &quot;Project Site&quot; column) and URL. One caveat here is that YAML does not allow tabs for whitespace. Indent with two spaces to remain consistent with the existing file format, please. Also note that the directory name is case-sensitive, as is always the case on Unix.


Finally, the HTML index file is generated from <code>index.mako</code>, a Mako template (which is mostly HTML anyhow). If you really can't figure out how it works, look up the Mako documentation.
Finally, the HTML index file is generated from <code>index.mako</code>, a Mako template (which is mostly HTML anyhow). If you really can't figure out how it works, look up the Mako documentation.
Line 147: Line 146:
# Create a zfs filesystem to store the project in:
# Create a zfs filesystem to store the project in:
#* Find the pool with least current disk usage
#* Find the pool with least current disk usage
#* <code>zfs create cscmirror{1,2}/$PROJECT_NAME</code>
#* <code>zfs create cscmirror{1,2,3}/$PROJECT_NAME</code>
# Change the folder ownership
# Change the folder ownership
#* <code>chown mirror:mirror /mirror/root/.cscmirror{1,2}/$PROJECT_NAME</code>
#* <code>chown mirror:mirror /mirror/root/.cscmirror{1,2,3}/$PROJECT_NAME</code>
# Create the symlink in <code>/mirror/root</code>
# Create the symlink in <code>/mirror/root</code>
#* <code>ln -s .cscmirror{1,2}/$PROJECT_NAME $PROJECT_NAME</code> ('''NOTE''': The symlink must be relative to the <code>/mirror/root</code> directory. If it isn’t, the symlinks will not work when chrooted)
#* <code>ln -s .cscmirror{1,2,3}/$PROJECT_NAME $PROJECT_NAME</code> ('''NOTE''': The symlink must be relative to the <code>/mirror/root</code> directory. If it isn’t, the symlinks will not work when chrooted)
# Repeat the above steps on mirror-dc. <code>sudo ssh mirror-dc</code> on potassium-benzoate
# Repeat the above steps on mirror-dc. <code>sudo ssh mirror-dc</code> on potassium-benzoate
# Configure the project in merlin (<code>~mirror/merlin/merlin.py</code>)
# Configure the project in merlin (<code>~mirror/merlin/merlin.py</code>)

Revision as of 17:19, 17 August 2021

The Computer Science Club runs a public mirror (mirror.csclub.uwaterloo.ca) on potassium-benzoate.

We are listed on the ResNet "don't count" list, so downloading from our mirror will not count against one's ResNet quota.

Software Mirrored

A list of current archives (and their respective disk usage) is listed on our mirror's homepage at mirror.csclub.uwaterloo.ca.

Mirroring Requests

Requests to mirror a particular distribution or archive should be made to syscom@csclub.uwaterloo.ca.

Implementation Details

Syncing

Storage

All of our projects are stored on one of three zfs zpools. There are 8 drives per array (7 run cscmirror3), configured as raidz2, and there is an additional drive that can be swapped in (in the event of a disk failure).

  • /mirror/root/.cscmirror1
  • /mirror/root/.cscmirror2
  • /mirror/root/.cscmirror3

Each project is given a filesystem under one of the two pools. Symlinks are created /mirror/root to point to the correct pool and file system.

Merlin

The synchronization process is run by a Python script called "merlin", written by a2brenna. The script is stored in ~mirror/merlin.

The list of repositories and their configuration (synch frequency, location, etc.) is configured in merlin.py.

To view the sync status, execute ~mirror/merlin/arthur.py status. To force the sync of a project, execute ~mirror/merlin/arthur.py sync:PROJECT_NAME.

Push Sync

Some projects support push syncing via SSH.

We are running a special SSHD instance on mirror.csclub.uwaterloo.ca:22. This instance has been locked down, with the following settings:

  • Only SSH key authentication
  • Only users of the push group (except mirror) are allowed to connect
  • X11 Forwarding, TCP Forwarding, Agent Forwarding, User RC and TTY are disabled
  • Users are chrooted to /mirror/merlin

Most projects will connect using the push user. The SSH authorized keys file is located at /home/push/.ssh/authorized_keys. An example entry is:

restrict,no-port-forwarding,no-X11-forwarding,no-agent-forwarding,no-pty,command="arthur sync:ubuntu >/dev/null 2>/dev/null </dev/null &",from="XXX.XXX.XXX.XXX" ssh-rsa ...

Sync Scripts

Our collection of synchronization scripts are located in ~mirror/bin. They currently include:

  • csc-sync-apache
  • csc-sync-debian
  • csc-sync-debian-cd
  • csc-sync-gentoo
  • csc-sync-ssh
  • csc-sync-standard

Most of these scripts take the following parameters:

local_dir rsync_host rsync_dir

HTTP(s)

We use nginx as our webserver.

Index

An index of the archives we mirror is available at mirror.csclub.uwaterloo.ca.

As of Winter 2010, it is now generated by a Python script in ~mirror/mirror-index.

~mirror/mirror-index/make-index is scheduled in /etc/cron.d/csc-mirror to be run hourly. The script can be run manually when needed (for example, when the archive list is updated) by running:

sudo -u mirror /home/mirror/mirror-index/make-index.py

The script will iterate all folders in /mirror/root, identify the size of the project using `zfs get -H -o value used $dataset`, where $dataset is calculated from the symlink in /mirror/root. The size of all folders is added together to calculate the total folder size (the total size includes hidden projects).

make-index.py is configured by means of a YAML file, config.yaml, in the same directory. Its format is as follows:

docroot: /mirror/root
output: /mirror/root/index.html

exclude:
   - include
   - lost+found
   - pub
# (...)

directories:
  apache:
    site: apache.org
    url: http://www.apache.org/

  archlinux:
    site: archlinux.org
    url: http://www.archlinux.org/

# (...)

The docroot is the directory which is to be scanned; this will probably always be the mirror root from which Apache serves. This is here so that it's easy to find and alter. For instance, we could change --human-readable to --si if we ever decided that, like hard disk manufacturers, we want sizes to appear larger than they are. output defines the file to which the generated index will be written.

exclude specifies the list of directories which will not be included in the generated index page (since, by default, all folders are included in the generated index page).

Finally, directories specifies the information of directories. All directories are listed by default, whether or not they appear in this list - only those under exclude are ignored. The format is fairly straightforward: simply name the directory and provide a site (the display name in the "Project Site" column) and URL. One caveat here is that YAML does not allow tabs for whitespace. Indent with two spaces to remain consistent with the existing file format, please. Also note that the directory name is case-sensitive, as is always the case on Unix.

Finally, the HTML index file is generated from index.mako, a Mako template (which is mostly HTML anyhow). If you really can't figure out how it works, look up the Mako documentation.

FTP

We use proftpd (standalone daemon) as our FTP server.

To increase performance, we disable DNS lookups in proftpd.conf:

UseReverseDNS           off
IdentLookups            off

We also limit the amount of CPU/memory resources used (e.g. to minimize Globbing resources):

RLimitCPU               session 10
RLimitMemory            session 4096K

We allow a maximum of 500 concurrent FTP sessions:

MaxInstances            500
MaxClients              500

The contents of /mirror/root/include/motd.msg are displayed when a user connects.

rsync

We use rsyncd (standalone daemon).

We disable compression and checksumming in rsyncd.conf:

dont compress = *
refuse options = c delete

The contents of /mirror/root/include/motd.msg are displayed when a user connects.

Mirror Administration

Adding a new project

  1. Find the instructions for mirroring the project. Ideally, try to sync directly from the project’s source repository.
    • Note that some projects provide sync scripts, however we generally won’t use them. We will instead use our custom ones.
  2. Create a zfs filesystem to store the project in:
    • Find the pool with least current disk usage
    • zfs create cscmirror{1,2,3}/$PROJECT_NAME
  3. Change the folder ownership
    • chown mirror:mirror /mirror/root/.cscmirror{1,2,3}/$PROJECT_NAME
  4. Create the symlink in /mirror/root
    • ln -s .cscmirror{1,2,3}/$PROJECT_NAME $PROJECT_NAME (NOTE: The symlink must be relative to the /mirror/root directory. If it isn’t, the symlinks will not work when chrooted)
  5. Repeat the above steps on mirror-dc. sudo ssh mirror-dc on potassium-benzoate
  6. Configure the project in merlin (~mirror/merlin/merlin.py)
    • Select the appropriate sync script (typically csc-sync-standard) and supply the appropriate parameters
  7. Restart merlin: systemctl restart merlin
  8. Configure the project in zfssync.yml (~mirror/merlin/zfssync.yml)
    • This will kick off the initial sync
    • Check ~mirror/merlin/logs/$PROJECT_NAME for errors, ~mirror/merlin/logs/transfer.log for transfer progress
  9. Update the mirror index configuration (~mirror/mirror-index/config.yaml)
  10. Add the project to rsync (/etc/rsyncd.conf)
    • Restart rsync with systemctl restart rsync

If push mirroring is available/required, see Push Sync.

Secondary Mirror

The School of Computer Science's CSCF has provided us with a secondary mirror machine located in DC. This will limit the downtime of mirror.csclub in the event of an outage affecting the MC machine room.

Keepalived

Mirror's IP addresses (129.97.134.71 and 2620:101:f000:4901:c5c::f:1055) have been configured has VRRP address on both machines. Keepalived does the monitoring and selecting of the active node.

Potassium-benzoate has higher priority and will typically be the active node. A node's priority is reduced when nginx, proftpd or rsync are not running. Potassium-benzoate starts with a score of 100 and mirror-dc starts with a priority of 90 (higher score wins).

When nginx is unavailable (checked w/ curl), the priority is reduced by 20. When proftpd is unavailable (checked with curl), the priority is reduced by 5. When rsync is unavailable (checking with rsync), the priority is reduced by 15.

The Systems Committee should received an email when the nodes swap position.

Project synchronization

Only potassium-benzoate is configure with merlin. mirror-dc has the software components, but they are probably not update to date nor configured to run correctly.

When a project sync is complete, merlin will kick off a custom script to sync the zfs dataset to the other node. These scripts live in /usr/local/bin and in ~mirror/merlin.