On Upcoming FOSDEM 2021

FOSDEM 2021 starts tomorrow. This time I have 3 talks to present. Here they are, in the order of appearance, with links to my related blog posts:

1. "Upgrading to a newer major version of MariaDB" - it should start at 10:30 on Saturday in the MariaDB devroom and is mostly devoted to mysql_upgrade. Related blog posts are:
   
   • "What mysql_upgrade really does in MariaDB, Part I " - original study of internal workings of mysql_upgrade utility inspired by Monty's post.
   • "What mysql_upgrade really does in MariaDB, Part II, Bugs and Missing Features" - recent review of bugs and feature requests for mysql_upgrade reported since the previous blog post.
     
2. "Monitoring MariaDB Server with bpftrace on Linux" - it should start at 12:40 on Sunday in the Monitoring and Observability devroom and is devoted to bpftrace basics. A lot more information is provided in recent blog posts:
   
   • "Playing with recent bpftrace and MariaDB 10.5 on Fedora - Part I, Basic uprobes" - few details on building recent version from source and basic probes to capture SQL queries and their execution time and to trace pthread_mutex_lock calls
   • "Playing with recent bpftrace and MariaDB 10.5 on Fedora - Part II, Using the Existing Tools" - review and tests of some bpftrace-based tools/small programs that come with it, mostly related to disk I/O monitoring. It's a;ways great to study by example and have more tools for typical tasks.
   • "Playing with recent bpftrace and MariaDB 10.5 on Fedora - Part III, Creating a New Tool for Tracing Mutexes" - on my first lame attempt to get interesting stacks and time spent waiting in them for pthread_mutex_lock and pthread_mutex_unlock pair of functions.
   • "Playing with recent bpftrace and MariaDB 10.5 on Fedora - Part IV, Creating a New Tool for Tracing Time Spent in __lll_lock_wait" - in that case I tried to measure time, per unique stack, spent inside the __lll_lock_wait function when MariaDB was under high concurrent load. Interesting results, but the performance drop is notable and it took too much time to get the outputs...
   • "Playing with recent bpftrace and MariaDB 10.5 on Fedora - Part V, Proper Way To Summarize Time Spent per Stack" - in this last post I finally managed to find a proper balance of spreading processing between bpftrace and user space awk etc tools, and got fast and useful results. Some more bpftrace features were also used.
     
3. "Linux /proc filesystem for MySQL DBAs" - this talk should start at 15:00 on Sunday in MySQL devroom and is devoted to a totally different (comparing to dynamic tracing I am so big fan of) way to get insights about MySQL internal working, waits etc - sampling of files in /proc file system. I had written 3 blog posts on the topic:
   
   • "Linux /proc Filesystem for MySQL DBAs - Part I, Basics" - mostly quotes from man 5 proc and few tests with Percona Server 5.7.x running under load.
   • "Linux /proc Filesystem for MySQL DBAs - Part II, Threads of the mysqld Process" - some basic ideas on how to match MySQL "thread"/connection with Linux thread that we can then sample via /proc. The approach works for MySQL 5.7+ and MariaDB 10.5+. For older versions one probably has to use gdb to identify threads for monitoring.
   • "Linux /proc Filesystem for MySQL DBAs - Part III, 0x.tools by Tanel Poder" - using these great and simple tools for regular monitoring and... to create an off-CPU flame graphs for MySQL server that is I/O bound or otherwise spe3nds time waiting on something in the kernel.
     

Slides are uploaded to the talks pages and will be shared via SlideShare. Draft, longer versions of the talks that I've recorded but FOSDEM system had not accepted will also be shared at my YouTube channel soon.

Usual view on my way to FOSDEM ULB site... I miss Brussels!

Other talks in these devrooms that I consider interesting (will not be able to attend them all live though):

• Migrating MariaDB Cluster to ARM
  
• Atomic DDL in MariaDB
• Buffer pool performance improvements 
• Performance Analysis and Troubleshooting Methodologies for Databases 
• Open Source Database Infrastructure with Vitess 
• Making MySQL-8.0 XA transaction processing crash safe 
• Rewrite Your Complex MySQL Queries for Better Performance
  

See you there! FOSDEM was a real driver of my more or less advanced performance studies this year.