Les outils linux pour savoir ce qui se passe.
http://www.brendangregg.com/Perf/linux_perf_tools_full.png

(via http://sebsauvage.net/links/?C6DJrw)

Performance User Group de Paris.

A lire plus tard

Top CPU par process :

top -d 1.0 -b | grep mysql

Top CPU per memory :

top -o %MEM -c -d 2

dtrace : disque et réseau
strace <programme> : tous les appels systèmes, mode debug, donc performance tombe à 0
opensnoop : moins gourmand que strace, se base sur le nom du process. Fichier script (bash) https://github.com/brendangregg/perf-tools/blob/master/opensnoop
netcat, netstat : réseau (pour netstat, option -tunapl)
ngrep, tcpdump : espion réseaux
perf : fonction système qui prenne du temps, pour faire des flamegraph

Thanks Julia !

(via http://sebsauvage.net/links/?5Kxo5A via Sam&Max)

et j'ajoute :

pmap <pid>

(via http://sametmax.com/verifier-la-ram-que-consome-un-process-sous-linux-avec-pmap/)

Et un exemple :

strace -o <file> -f -e trace=all -p <pid>

Grapher les sorties vmstat

– durability setting ( innodb_flush_log_at_trx_commit = 0 or 1 ) ?
– do we use binary logs ( I used ROW based replication for 5.1)
– do we have sync_binlog options.

• innodb size :
  https://www.percona.com/blog/2015/06/02/80-ram-tune-innodb_buffer_pool_size/

Utilisation CPU d'une JVM

 -XX:+PreserveFramePointer

Bientôt dans toutes les bonnes JVM

Tout est dit : sar grapher

/coude (via https://geekz0ne.fr/shaarli/?kBiReg)

Résumé des commandes :

uptime
dmesg | tail
vmstat 1
mpstat -P ALL 1
pidstat 1
iostat -xz 1
free -m
sar -n DEV 1
sar -n TCP,ETCP 1
top

(via https://jeekajoo.eu/links/?hAhfXA)

jmeter performance tuning

Performance SQL

mysql / mariadb performance tip

iotop
dstat
while true; do date; ps auxf | awk '{if($8=="D") print $0;}'; sleep 1; done

quel outil pour observer linux (via http://sebsauvage.net/links/?FdQLWg)

pour afficher les informations qui sont tracées dans le schéma de performance.

select * from setup_consumers; 

events_stages_current   NO
events_stages_history   NO
events_stages_history_long  NO
events_statements_current   YES
events_statements_history   NO
events_statements_history_long  NO
events_waits_current    NO
events_waits_history    NO
events_waits_history_long   NO
global_instrumentation  YES
thread_instrumentation  YES
statements_digest   YES

Vider les tables du schéma de performances :

truncate table events_statements_current; 
truncate table events_statements_history; 
truncate table events_statements_summary_by_digest;

• dans la page, affichage des requêtes pendant 60 secondes :

use performance_schema;
update setup_consumers set enabled='YES' where name IN ('events_statements_history','events_statements_current','statements_digest');
truncate table events_statements_current; truncate table events_statements_history; truncate table events_statements_summary_by_digest;
do sleep(60);
select now(),(count_star/(select sum(count_star) FROM events_statements_summary_by_digest) * 100) as pct, count_star, left(digest_text,150) as stmt, digest from events_statements_summary_by_digest order by 2 desc;
update setup_consumers set enabled='NO' where name IN ('events_statements_history','events_statements_current','statements_digest');

Un ensemble de requêtes pour utiliser le schéma de perf :
http://www.markleith.co.uk/ps_helper/#statements_with_runtimes_in_95th_percentile
http://www.markleith.co.uk/ps_helper/#statements_with_full_table_scans

Pour les requêtes :

SELECT * FROM performance_schema.events_statements_summary_by_digest;

Top par SUM :

SELECT schema_name, digest, digest_text, count_star, round(AVG_TIMER_WAIT/1000000000), sum_errors, sum_warnings, sum_rows_affected, sum_rows_sent, sum_rows_examined FROM `events_statements_summary_by_digest` ORDER BY `events_statements_summary_by_digest`.`SUM_TIMER_WAIT`  DESC

Top par SUM avec Full Scan :

SELECT DIGEST_TEXT AS query,
      IF(SUM_NO_GOOD_INDEX_USED > 0 OR SUM_NO_INDEX_USED > 0, '*', '') AS full_scan,
      COUNT_STAR AS exec_count,
      SUM_ERRORS AS err_count,
      SUM_WARNINGS AS warn_count,
      ROUND(SEC_TO_TIME(SUM_TIMER_WAIT/1000000000000)) AS exec_time_total_second,
      ROUND(SEC_TO_TIME(MAX_TIMER_WAIT/1000000000000)) AS exec_time_max_second,
      ROUND((AVG_TIMER_WAIT/1000000000)) AS exec_time_avg_ms,
      SUM_ROWS_SENT AS rows_sent,
      ROUND(SUM_ROWS_SENT / COUNT_STAR) AS rows_sent_avg,
      SUM_ROWS_EXAMINED AS rows_scanned,
      DIGEST AS digest
FROM performance_schema.events_statements_summary_by_digest ORDER BY SUM_TIMER_WAIT DESC LIMIT 15

des hints pour la performance :

• index inutilisés (SELECT object_schema, object_name, index_name FROM performance_schema.table_io_waits_summary_by_index_usage WHERE index_name IS NOT NULL AND count_star = 0 ORDER BY object_schema, object_name;)
• table temporaire

Intéressant : select * from file_summary_by_instance order by count_read desc,count_write desc limit 5