Accidental DBA

Upgrading Evergreen: A Case Study

Managing a Software Upgrade

Discoveries in Digital Humanities: How Evergreen Contributes

Accidental DBA

Josh Berkus (with help from Anoop Atre) -

Slides are here.

Introductory Comments

Reference documentation:

Key Terms/Touch Words/Slang

Changing Playing Field


Package Types:

Solaris Packages exist -- longer list in slides.

Data Directories

Major interfaces -- Command Line

Evergreen Extensions

Install Versions

psql -f hstore.sql


Major update announcements -- important security update happened last week.

Major Upgrades -- once a year

Upgrading Extensions


Host-based access

These are compliant with IPB6 (but no IPB6 examples included).

Security Tips


Have a DR plan!

Binary replication




Example demo on replication

Consideration during IRL implementations

Failover vs. load-balancing

There are more complicated replication methods


OS Check -- things to monitor and trend

Database Checks

See slides for others

Activity Log

Garbage Collection

You can check PGSTAT USER TABLES to check for table bloat.

Fight that uncomfortable bloating

XID Wraparound

No time for Bonus Round time, straight to questions:

Q: When you talk about replication that can be used for failover or for load balancing, could you have two replica servers?

A: For users with high requirements, yes. The reporting server can be used for failover, just not rapid failover. When you failover, it will notice that it is 3 hours behind and prevent it from going ahead. I didn’t explain synchronous replication -- this is for high needs users.

Synchronous Replication people will also have a synch replicas.

Q: Features List of in 9.2 features that are key

A:Vastly better performance for synchronous environments, Limited set of Jsong features, better performance

Q: When was auto vacuum added?

A: 8.3

Q: What you do in a DB with connections set to 1000?

A: Well, you are wasting memory tracking these. Depending on OS, if you actually used more than half the connections, you will see an actual order magnitude drop in efficiency; there is an issues of process accounting. Back in the Linux mis 2 series, when you hit a lot of processes, it would switch processes and efficiency would drop. It will drop a ton.

Q: What recommended Linux scheduler do you use?

A: This used to be a big deal. they changed it. The 3 series should be left at the default unless I have good storage in which case I use in no I?

They changed the algor and it is now working fine in post-res.

Upgrading Evergreen: A Case Study

Jeff Davis

^        |

|        v



Managing a Software Upgrade

Shauna Borger

Discoveries in Digital Humanities: How Evergreen Contributes

Vyacheslav Tykhonov \ Mieke Stroo - Evergreen tools at IISH

Suggestion: Evergreen community send a committee to work over IISH for an extended period of time.

A: Need a Claim module for Serials. Browse searching for call numbers. (Currently binary search)