Hi Does anyone have any experience or advice about running two servers on the same database?
The pros are that I can test tools and some other changes against actual datasets and check bug fixes directly by giving users temporary access to a dev server so that they regress the bugs themselves. The cons are that I risk damaging user data and ending up with a pile of rubble.
I would like to have a production server and development server both running simultaneously. There are obvious issues such as the schema needs to be compatible for both systems simultaneously. This might mean disabling or enhancing the schema version control mechanism.
What else needs to be modified to make this work?
As the wicked witch of the west said "these things must be done very del-i-cate-ly" ;-)
Thanks, Ted
On Mon, Mar 25, 2013 at 5:41 PM, Ted Goldstein tedgoldstein@gmail.com wrote:
Hi Does anyone have any experience or advice about running two servers on the same database?
My advice is don't do it. The moment you update the development server to run with a new database schema version, your shared database will be updated, and the production server will probably stop working. That's assuming there are not other issues before that...
(If you are talking about running two version-locked production servers, then perhaps there is precedent with load balancing systems which might be worth investigating)
Peter
Hello Ted!
Peter Cock wrote, On 03/25/2013 01:51 PM:
On Mon, Mar 25, 2013 at 5:41 PM, Ted Goldstein tedgoldstein@gmail.com wrote:
Hi Does anyone have any experience or advice about running two servers on the same database?
My advice is don't do it.
I'll second Peter's advice, don't ever do that :)
I forgot where I heard it, but someone (wise) said "Code flows from dev to prod, data should flow from prod to dev".
To facilitate that, we use the following script, that can quickly copy the data from our production server to our development server. Once it's run, the databases contain the same datasets and workflows and everything, making testing and debugging much easier:
=== #!/bin/sh
DATE=$(date "+%Y_%m_%d_%H%M%S")
FILE="galaxy_db_prod_to_devel_${DATE}.sql.gz"
DEST=/tmp/ FILE="${DEST}${FILE}"
echo "Dumping Prod-DB to:" echo " $FILE" echo "(Enter 'galaxyprod' DB password)" pg_dump -c -U galaxyprod galaxyprod | sed 's/galaxyprod/galaxydev/g' | gzip > "$FILE" || exit 1
echo echo "Dropping and re-creating Galaxy-Devel database." echo "press CTRL-C to abort or ENTER to continue." read ## First, drop all tables, functions, etc.
psql -t -d galaxydev -c "SELECT 'DROP TABLE ' || n.nspname || '.' ||c.relname || ' CASCADE;' FROM pg_catalog.pg_class AS c LEFT JOIN pg_catalog.pg_namespace AS n ON n.oid = c.relnamespace WHERE relkind = 'r' AND n.nspname NOT IN ('pg_catalog', 'pg_toast') AND pg_catalog.pg_table_is_visible(c.oid)" | psql galaxydev zcat "$FILE" | psql galaxydev || exit 1
## Reset all the waiting jobs, we don't want to re-run them on dev psql galaxydev -c "update job set state='error' where (state='new' or state='waiting');"
rm "$FILE" ===
Once this script is complete, the "galaxydev" database is an exact copy of "galaxyprod" database (except queued jobs).
Hope this helps, -gordon
Couple of clarifications for the script, which aren't obvious:
Assaf Gordon wrote, On 03/25/2013 02:03 PM:
To facilitate that, we use the following script, that can quickly copy the data from our production server to our development server. Once it's run, the databases contain the same datasets and workflows and everything, making testing and debugging much easier:
=== #!/bin/sh
DATE=$(date "+%Y_%m_%d_%H%M%S")
FILE="galaxy_db_prod_to_devel_${DATE}.sql.gz"
DEST=/tmp/ FILE="${DEST}${FILE}"
echo "Dumping Prod-DB to:" echo " $FILE" echo "(Enter 'galaxyprod' DB password)" pg_dump -c -U galaxyprod galaxyprod | sed 's/galaxyprod/galaxydev/g' | gzip > "$FILE" || exit 1
echo echo "Dropping and re-creating Galaxy-Devel database." echo "press CTRL-C to abort or ENTER to continue." read ## First, drop all tables, functions, etc.
psql -t -d galaxydev -c "SELECT 'DROP TABLE ' || n.nspname || '.' ||c.relname || ' CASCADE;' FROM pg_catalog.pg_class AS c LEFT JOIN pg_catalog.pg_namespace AS n ON n.oid = c.relnamespace WHERE relkind = 'r' AND n.nspname NOT IN ('pg_catalog', 'pg_toast') AND pg_catalog.pg_table_is_visible(c.oid)" | psql galaxydev zcat "$FILE" | psql galaxydev || exit 1
## Reset all the waiting jobs, we don't want to re-run them on dev psql galaxydev -c "update job set state='error' where (state='new' or state='waiting');"
rm "$FILE"
Database "galaxyprod" with user "galaxyprod" are configured to ask for a password, which is why: pg_dump -c -U galaxyprod galaxyprod Will stop and ask for a password. Local unix users can't automatically access "galaxyprod".
Database "galaxydev" is configured to be accessible to few local unix users, which is why: psql galaxydev Doesn't need an explicit user name and won't ask for a password (and is run multiple times in the script).
This is configured in "/var/lib/pgsql/data/pg_hba.conf" as: === # TYPE DATABASE USER CIDR-ADDRESS METHOD local galaxydev gordon trust ===
Whether it's secure/optimal or not is open for discussion...
The convoluted SQL is used to drop all tables before adding the new ones. It's based on this message: http://www.postgresql.org/message-id/1093378065.15248.36.camel@linda
With this script we also test database upgrades without worries, as we first copy the prod database to dev, then run "sh manage_db.sh upgrade" on the development database. If it works - great. If not (rare, but happened once or twice), we do not upgrade prod.
-gordon
galaxy-dev@lists.galaxyproject.org
-
Assaf Gordon -
Peter Cock -
Ted Goldstein