install.rstx

=========================================
GAVO DaCHS installation and configuration
=========================================

:Author: Markus Demleitner
:Email: gavo@ari.uni-heidelberg.de
:Date: |date|
:Copyright: Waived under `CC-0`_


.. contents::
  :depth: 3
  :backlinks: entry
  :class: toc

These installation instructions cover the installation of the complete
data center suite.  Installing libraries or, say, the tapsh, is much
less involved.  See the respective pages at the `GAVO DC's software
distribution pages`_ for details on those.

.. _GAVO DC's software distribution pages: http://vo.ari.uni-heidelberg.de/soft


Debian Systems (and Derivatives)
================================

The preferred way to run DaCHS is on Debian stable or compatible
systems.   Starting with Debian bullseye (11.0), DaCHS is part of
Debian.  On a suitably recent system, just execute::

  sudo apt install gavodachs2-server

With that, you are ready to proceed to the tutorial_.

This is the recommended way to get up to speed with DaCHS regardless of
whether you want to run the bleeding-edge version.  You can
:dachsdoc:`switch later <howDoI.html#switch-between-svn-and-packaged-dachs>`.

On older systems, you need to `add our APT repository`_ to your
``/etc/apt/sources.list`` before running the ``apt install``.

You are probably saving yourself quite a bit of grief if you just choose
Debian stable as your platform when you are going for Debian
derivatives.  *If* you insist on running, say, Ubuntu, please note that
they do not package DaCHS (so you have to add our APT repo no matter
what) and that they do not package the q3c and pgsphere packages that
DaCHS cannot live without.

A workable fix is to install the packages from postgres' APT repository
(these are basically equivalent to what Debian has).  However, these
have multiple postgres versions at the same time, and hence you will
need to manually make sure that the postgres server and the extensions
match.

Hence, on dpkg-based distributions more remote from Debian, try a
sequence more or less like:

(#) Purge all postgres packages that may already be on your system
(#) `Add the postgres APT repository`_
(#) `Add our APT repository`_
(#) ``sudo apt update``
(#) ``export PGVER=13; sudo apt install postgresql-$PGVER postgresql-$PGVER-q3c postgresql-$PGVER-pgsphere``
(#) ``apt install gavodachs2-server``

.. _Add the postgres apt repository: https://www.postgresql.org/download/linux/debian/

If this fails, make sure the postgres instance listening on your port
5432 really *is* the one you got from the postgres repo, *not* the one
from Ubuntu.

.. _add our apt repository: http://soft.g-vo.org/repo
.. _tutorial: tutorial.html
.. _Operator's Guide: opguide.html


RPM-based Distributions
========================

The following instructions were mainly written by Chris Piker (U Iowa)
– thanks! – and reflect the state on CentOS 7.  The systemd parts are
probably applicable beyond the RPM universe.


Setup PostgreSQL with an alternate database cluster using the C locale
----------------------------------------------------------------------

Install postgres if needed::

  $ sudo yum install postgresql-server postgresql-devel

If postgres is running shut it down::

  $ sudo systemctl stop postgresql.service

Moving the database cluster:  The database cluster is typically in the
``$HOME/data`` directory of the postgres user.  Changing this under CentOS 7
which uses systemd is a bit different.  Configuration files are now .ini
style files (not bash scripts) and they live under
``/usr/lib/systemd/system/``.

The default configuration file for postgres is
``/usr/lib/systemd/system/postgresql.service``.

The files in ``/usr/lib/systemd`` are maintained by the package manager so you
should leave them alone.  Forturnatly, systemd looks for overrides in an
alternate directory first before loading the deflault service configuration.

Local user defined overrides for services go in the
``/etc/systemd/system`` directory.  So, put::

     # Include the default config:
     .include /usr/lib/systemd/system/postgresql.service

     [Service]
     Environment=PGDATA=<your path here>/data
	
into ``/etc/systemd/system/postgresql.service``; of course *<your path
here>* needs to be substituted with the path to somewhere where there's
enough space for what you expect in your database; it's usually a good
idea to use a directory on a separate partition here.  The examples
below assume ``/disk/1/pgsql`` (and need to be adjusted accordingly in
the likey event that you chose something else).

Let systemd know that you've changed something::

  $ sudo /bin/systemctl daemon-reload

Also, the home directory for the postgres user should be changed to::

  $ $EDITOR /etc/passwd  (change postgres user home directory)
  $ sudo mkdir /disk/1/pgsql
  $ sudo cp /var/lib/pgsql/.bash_profile /disk/1/pgsql
  $ sudo chown -R postgres:postgres /disk/1/pgsql

Now initialize the db-cluster::

  $ sudo su -l postgres
  $ initdb -D $HOME/data --locale=C --lc-collate=C --lc-ctype=C -E UTF8
  $ $EDITOR data/pg_hba.conf
    > change 'trust' authentication to 'ident'
	
Start up the service::

  $ sudo systemctl enable postgresql.service
  $ sudo systemctl start postgresql.service

Test the service::

  $ sudo su -l postgres
  postgres$ psql
  postgres# \l

Make sure postgresql account has a password and will listen on the loop-back
interface::

  $ sudo su -l postgres
  $ psql
  postgres# alter user postgres encrypted password '*********'; (remember this)
  postgres# \q
  $ rm .psql_history  (or else the password is stored on disk)

  $ cd data
  $ vim pg_hba.conf
    <<< change <<<
       host    all         all             127.0.0.1/32            ident
	    host    all         all             ::1/128                 ident
    >>> to >>>>
       host    all         all             127.0.0.1/32            md5
	    host    all         all             ::1/128                 md5

and restart the postmaster::

  $ exit   (postgres account)
  $ sudo systemctl restart postgresql.service

Test that it works, you should get a password prompt::

  $ sudo -u postgres psql -h 127.0.0.1 -U postgres template1


Install distribution supplied dependency packages
-------------------------------------------------

This is realatively straighforward::

  $ sudo yum install python3-twisted #has lots of dependencies, yum gets them
  $ sudo yum install python3-pyparsing
  $ sudo yum install python3-astropy
  $ sudo yum install python3-matplotlib
  $ sudo yum install python3-setuptools
  $ sudo yum install python3-docutils
  $ sudo yum install python3-pillow  #New name for Python Image Manipulation Lib

Make/install extra packages
---------------------------

The RedHat/CentOS package universe is smaller than that available to Fedora
and Debian.  Instead of searching for rpms that may or may not be compatible
with your OS, is ususally more reliable to just build packages from source
if they are not available in the primary repositories.


PGSphere
''''''''

::

  $ git clone https://github.com/akorotkov/pgsphere.git
  $ cd pgsphere
  $ make USE_PGXS=1 PG_CONFIG=/usr/bin/pg_config
  $ sudo make USE_PGXS=1 PG_CONFIG=/usr/bin/pg_config install

Test (difficult due to user permissions, but possible)::

  $ mkdir results
  $ sudo chmod 777 . results
  $ sudo su postgres
  $ make USE_PGXS=1 installcheck
  $ exit
  $ rm regression.*
  $ sudo su postgres
  $ make USE_PGXS=1 crushtest
  $ exit


q3c
'''

::

  $ git clone https://github.com/segasai/q3c.git
  $ cd q3c
  $ make
  $ sudo make install



Automat
'''''''

There are finite-state machines for python, need by twisted. Version
0.3.0 is the newest that will work with the system supplied python
(Python 2.7.5)::

  $ wget https://github.com/glyph/automat/archive/v0.3.0.tar.gz
  $ tar -xvzf v0.3.0.tar.gz
  $ $EDITOR setup.py         # (look for potential issues)
  $ sudo python setup.py install --dry-run #(look for potential issues)
  $ sudo python setup.py install



Download/Patch/Build/Install GAVO DaCHS Software
------------------------------------------------

Getting the source is an easy download; these instructions were tested
with SVN revision 6511::

  $ svn http://svn.ari.uni-heidelberg.de/svn/gavo/python/trunk dachs
  $ cd dachs


Install editable, not touching any of the system packages::

  $ sudo -H pip install --no-deps --break-system-packages --no-build-isolation -e .
	
From here follow the standard instructions to test dachs.  With this,
you should be all set to try out stuff from :dachsdoc:`tutorial.html`.

The final step is to make DaCHS start as part of the system boot.


Setup DaCHS web-server as a system service
------------------------------------------

Use the following systemd unit file (put it into
``/etc/systemd/system/dachs.service``)::

  [Unit]
  Description=The DaCHS VO server
  After=network.target

  [Service]
  ExecStart=/usr/local/bin/dachs serve start -f
  Type=simple
  Restart=on-abnormal

  [Install]
  WantedBy=multi-user.target


You may need to adapt the path to the ``dachs`` binary.

Before activating DaCHS through systemd, make sure you have shutdown any
instances of dachs that you may have running::

  $ sudo su -l gavo
  $ dachs serve stop
  $ exit

Let systemd know a config file has changed::

  $ sudo /bin/systemctl daemon-reload

Then enable and start the service::

  $ sudo systemctl enable dachs.service
  $ sudo systemctl start dachs.service
  $ sudo systemctl status dachs.service



Installation without Package Management
=======================================

Dependencies
------------

Unfortunately, DaCHS has quite a few dependencies;
here's the list of dependencies of our Debian package as of version
2.1. This should give you some clue as to what might be necessary on
other systems::

  python3-astropy, python3-docutils, python3-numpy, python3-pil,
  python3-pkg-resources, python3-psycopg2, python3-pyparsing,
  python3-cryptography, python3-docutils, python3-lxml,
  python3-matplotlib, python3-pkg-resources, python3-psycopg2,
  python3-pymoc, python3-rjsmin, python3-testresources, python3-twisted,
  postgresql-q3c, postgresql-pgsphere

One thing you could also try is comment in the ``install_requires`` line
in DaCHS' setup.py (see below) and try to pull in the dependencies via
pip.  Expect minor breakage in that case, though, as we don't do
versioned dependencies there, and as libraries develop, their APIs do
change.

If you want to use boosters, you will additionally need::

  build-essential libcfitsio3-dev

To install from our version control system (see below), you will also
need::

  subversion

Of course, you'll need postgres itself on top of that.  We currently
require postgres 9.6 or newer.  If you actually need support for older
Postgres releases, let us know – it's not  hard to restore.


PgSphere
''''''''

PgSphere is a postgres extension for spherical geometry.  It is needed
for support of the geometric types in DaCHS' ADQL implementation and in
the preferred SIAP backend, so you should definitely install it.
So build it, install the server development packages for postgres (such
as postgresql-server-dev-9.x or postgresql-devel), check out
https://github.com/pgsphere/pgsphere, and in the source directory run::

  USE_PGXS=1 make
  sudo USE_PGXS=1 make install


Q3C
'''

DaCHS uses the Q3C library by Sergey Koposov and Oleg Bartunov,
http://www.sai.msu.su/~megera/oddmuse/index.cgi/SkyPixelization for
positional indexes.  DaCHS uses it for positional indexes (the
scs#q3cindex mixin) and in the interpretation of ADQL.  It is therefore
highly recommended to install it.

To do that, get the source directly
from https://github.com/segasai/q3c/releases/, install the server development
packages for postgres (such as postgresql-server-dev-9.x or postgresql-devel),
and in the source directory run::

  make
  sudo make install




Installing DaCHS
----------------


Getting the source
''''''''''''''''''

If you cannot use the Debian package (or do not want to), you can grab
a gavodachs package from `our distribution page`_.  Choose whatever
`gavodachs-latest.tar.gz` points to.

If you want to follow the bleeding edge closely – DaCHS is being
actively developed – check out whatever is in the subversion repository
right now.  For a read-only copy, say::

  svn co http://svn.ari.uni-heidelberg.de/svn/gavo/python/trunk/ dachs

After that, the current source code is in the ``dachs`` subdirectory.
This is development code, so *please* do not hesitate to contact us if
something weird is going on with it.  We mean it; even trivial reports
help us to gauge where our software behaves contrary to expectations.
Plus, we don't have oodles of users, so chances are you won't
get on our nerves.  For contact options see
http://docs.g-vo.org/DaCHS/#support.

.. _our distribution page: http://vo.ari.uni-heidelberg.de/soft/dist


Installing from source
''''''''''''''''''''''

The DaCHS installer is based on setuptools; we do not use setuptools'
dependency management, though, since in practice it seems more trouble
than it's worth, which means you need to manually install Dependencies_.

To install the software, in the ``dachs`` directory you checked out
above, say::

  sudo -H pip install --no-deps --break-system-packages --no-build-isolation -e .

(in DaCHS < 2.8.1, you should instead simply say ``sudo setup.py
develop``).  This looks ghastly and looks dangerous but is actually what
you need to do if you want to install a single package system-wide
without stepping on the distribution's packages and downloading code
from pypi.

This installs in “editable mode“, that is, if you edit anything in your
checkout, it will be reflected in what the system executes.  There is
nothing wrong with leaving out the ``-e`` flag, except that probably
defeats the purpose of installing from SVN in the first place.

**Note**: If running from SVN, do not forget to run ``dachs upgrade``
*after* an ``svn update``.  The on-disk structures of DaCHS sometimes
change, and ``dachs upgrade`` makes sure they are properly updated if
necessary.  Technically, you would only need to run gavo upgrade if, in
gavo --version, the two numbers behind “Schema” are different, but since
``dachs upgrade`` is smart enough to figure out when there's no need to
do anything, just make it a routine to run it.

**Note**: ``python setup.py install`` and friends do not install DaCHS'
man page.  Either do that manually (it's in docs/dachs.1) or use the
online version at :dachsdoc:`dachs.1.html`


Setup
=====

All this is taken care of by the Debian package, so don't do any of this
if you installed from .deb.


Introduction
------------

GAVO DaCHS is quite sensitive to a correct setup as regards permissions.
Experience has shown that user setup is the number one reason for
installation problems.  So, up front, here's what the steps given
below should create:

* A group  that will own certain directories that must be writable by
  the server (by default gavo).
* A user  that the server will run as (by default gavo).
* A unix account for you that should not be root (in particular not if
  you're using setup.py develop on an SVN checkout).  This should be in
  the gavo group (for when you're running ``dachs serve debug``) and will
  usually own resource directories and the like.

On the database side, the following must be ascertained:

* There's a postgres database cluster in the C locale, with a database
  already created (named, by default gavo).
* "you" (i.e., your unix id) have admin privileges on this (at least
  for installation) using ident authentication
* for connections from the local host, the three roles the server use
  can access the database using md5 authentication.


Account Management
------------------

You should first create a user that the DaCHS server runs as later,
and a group for running DC-related processes in::

  sudo adduser --system gavo
  sudo addgroup --system gavo
  sudo adduser gavo gavo

(or similar, depending on your environment). This user should not be
able to log in, but it should have a home directory.  Everyone that may
issue a ``dachs serve debug`` must be in the group created (this is
because the log directory will be writable by this group); in
particular, you should add yourself::

  sudo adduser `id -nu` gavo

You may want to create another account for "maintenance", or just use
your normal account; if more than one person will feed the data center,
you'll need more elaborate schemes.

To update the system's idea of your group membership, say ``newgrp
gavo`` or log in and out now.

All users that are to ingest data into the database using DaCHS must be
part of this gavo group.


Database setup
--------------

The most complicated step in setting up DaCHS is actually setting up the
database.  We currently only support postgres.

While it is conceivable to use DaCHS together with an existing postgres
database, we do not recommend trying this the first time.  Experiment
with a database dedicated to DaCHS first, then consider whether it's
worth interfacing to your existing database or whether a copy of that
data is more convenient.


Cluster Creation
''''''''''''''''

You first need a database to play with, preferably in a suitable
cluster (you could skip this, but the all bets are off as to whether
you'll be able to store non-ASCII characters in strings).

It is recommended to create a dedicated cluster first even if you want
to connect DaCHS to a pre-existing database later to get a feeling for how
it works.  See `Connecting to a remote database`_ for information on
what setup is necessary in this case.

Database cluster generator is very system-dependent, and ideally a
database admin would assist you.

On Debian systems dedicated for GAVO DaCHS, you can try the following
(*Warning*: This will destroy any previous content anyone put in
postgres databases on that particular system):

(#) Find out the version of the server you will be running (e.g., using
    ``dpkg -l``; in Debian, more than one version may be installed in
    parallel.  It's probably a good idea to use the most recent one.
    Set your desired version for subsequent use::

      export PGVERSION=11

(#) Drop the Debian default cluster (this will delete everything in
    there -- for a fresh install, that doesn't matter, but don't do this
    if other people use the database).  If you don't do this, your
    database will listen do a different port, and you will have to
    adapt the default profiles::

      sudo pg_dropcluster --stop $PGVERSION main

(#) Create the new cluster used by DaCHS::

      sudo pg_createcluster -d /<path-to-where-your-db-should-reside> \
        --locale=C -e UNICODE\
        --lc-collate=C --lc-ctype=C $PGVERSION main

    The locale should currently be C, because only the C locale will
    allow you databases with all kinds of encodings.  The database
    stores descriptions and similar entities, and you may encounter
    funny characters in there.  It would be a shame if you couldn't
    store them (plus, you would get odd error messages for those).

    If unsure where to put the cluster: Debian's default is
    ``/var/lib/postgresql/<postgres-version>/main``.

(#) Start the server::

       sudo /etc/init.d/postgresql start

(#) Create the database itself::

       sudo -u postgres createdb -Ttemplate0 --encoding=UTF-8 --locale=C gavo

On Debian, the configuration files for this cluster are at
``/etc/postgresql/$PGVERSION/pgdata/``.



Initial Account Setup
'''''''''''''''''''''

At least during setup, you also need superuser privileges on the
database.  For ``dachs init`` below to work, your normal account must
have such privileges.  On Debian systems, you can simply say::

  sudo -u postgres createuser -s `id -nu`

You can drop those privileges later if they make you nervous, but for
gavo init you need to be DB superuser.  Also note that DaCHS assumes
your server is trusted, and if people have managed to take over an
account in the gavo group, they can do with your database whatever they
please anyway.  In particular (don't complain we didn't tell you), DaCHS
currently encrypts *no* passwords; for the DB passwords, sensible
encryption would mean the software requires some passphrase during
startup, which we don't want.  For user passwords (for protecting web
resources), it would make no sense since with HTTP basic authentication
as employed by DaCHS, they travel through the net unencrypted anyway
(which is sometimes called "mild security").


Connecting to a Remote Database
'''''''''''''''''''''''''''''''

See :dachsdoc:`opguide.html#two-server-operation`.


Configuration File
------------------

Next, you need to decide on a "root" directory for DaCHS.  Below it,
there are data descriptions, cache files, logs, etc. (these locations
can be changed later, but for a simple setup we recommend keeping
everything together).  By default, this is ``/var/gavo``.

DaCHS is configured in an INI-style configuration file in
``/etc/gavo.rc`` (overridable using the envirvonment variable
``GAVOSETTINGS``).  In addition, users, in particular the gavo user, can
have ~/.gavorc files, the contents of which override settings in
/etc/gavo.rc.

`Configuration Settings <./opguide.html#configuration-settings>`_
gives a walkthrough through the most important
settings; for now, you must set the DaCHS root dir if you are not happy
with ``/var/gavo``::

  [general]
  rootDir: /data/gavo

as /etc/gavo.rc.

Whatever ``rootDir`` is, it must exist and be writable by you, or you
must have sufficient privileges to create it.  Do *not* run ``dachs
init`` as root, since the files and directories it creates will be owned
by whoever ran the program.  In the typical situation in which you may
not write to ``rootDir``'s parent, do something like::

  sudo mkdir -p /data/gavo
  sudo chown `id -nu`:gavo /data/gavo

You can now let DaCHS create its file system hierarchy::

  dachs init

``dachs init`` will spit out a warning about a missing file
``defaultmeta.txt`` on the first run.  On that first run, you can ignore
the warning; the missing file will be created by DaCHS.

If your database server is not on the same machine as your web server
(which is not recommended for a test setup), you have to pass a complete
DSN that lets DaCHS connect as a superuser to ``dachs init``.  A DSN
("Data Source Name") is a sequence of key-values pairs as used  by ODBC
or Postgres itself (with keys discussed in `Database Connection Control
Functions in the postgres documentation`_).  You would say something
like::

  dachs init --dsn "host=myhost.xy port=5546 user=super password=secret dbname=wisdom"

– make sure you give at least dbname and whatever role DaCHS ends up
using has superuser privileges during setup (that role is not used
during normal DaCHS operation any more).

You can later run gavo init again. It will not clobber anything you did
in the meantime (well, if it does, it's a bug and you should fiercely
complain).  In particular, this is the most convenient way to create
directories if you changed locations in ``gavo.rc``.

.. _Database Connection Control Functions in the postgres documentation: http://www.postgresql.org/docs/current/static/libpq-connect.html


The init script
===============

Though you can operate the server manually through ``dachs serve
(start|stop|reload)``, you will probably want to have your init system
automatically start it.  See
https://docs.g-vo.org/DaCHS/tutorial.html#starting-and-stopping-the-server
for details on integrating DaCHS with either sysvinit or systemd.


Docker
======

**We do not really recommend using Docker containers on production
machines.**  We do, however, use Docker in our internal QA to check
installability and upgradeability; in case you are curious what we do,
see http://svn.ari.uni-heidelberg.de/svn/integration/dockerbased .

Still, if you want to build a Docker-able DaCHS take a look at the
corresponding `Git <https://github.com/chbrandt/docker-dachs>`_
or, analogously, `Docker <https://hub.docker.com/r/chbrandt/dachs/>`_
repositories.

.. |date| date::
.. _CC-0: http://creativecommons.org/publicdomain/zero/1.0