For several different EventLogging tables, the same query yields quite different results depending on whether it is run on db1047 (= "s1-analytics-slave") or on dbstore1002 ("analytics-store" = "s2-analytics-slave" etc.).
Example 1 (about MobileWebSectionUsage_15038458, queries done within a few hours on Saturday/Sunday March 26/27):
see P2832
Example 2: (both queries ran just now, and earlier on March 16 with the same results)
mysql:research@analytics-store.eqiad.wmnet [(none)]> SELECT COUNT(*) AS events FROM log.NavigationTiming_14899847 WHERE LEFT(timestamp, 8) = '20151203'; +--------+ | events | +--------+ | 307440 | +--------+ 1 row in set (5.84 sec) mysql:research@s1-analytics-slave.eqiad.wmnet [(none)]> SELECT COUNT(*) AS events FROM log.NavigationTiming_14899847 WHERE LEFT(timestamp, 8) = '20151203'; +--------+ | events | +--------+ | 358908 | +--------+ 1 row in set (2 min 15.64 sec)
Example 3: Many tables seem to differ considerably in size between store and s1, e.g. these (queried just now):
mysql:research@s1-analytics-slave.eqiad.wmnet [(none)]> SELECT table_name, (DATA_LENGTH + INDEX_LENGTH)/1024/1024/1024 as `TOTAL SIZE (GB)` FROM information_schema.tables WHERE table_name LIKE 'MobileWikiAppToC%'; +--------------------------------------+------------------+ | table_name | TOTAL SIZE (GB) | +--------------------------------------+------------------+ | MobileWikiAppToCInteraction_10375484 | 150.709785191342 | | MobileWikiAppToCInteraction_11014396 | 0.988343123347 | | MobileWikiAppToCInteraction_14585319 | 0.452323913574 | | MobileWikiAppToCInteraction_8461467 | 81.039179009385 | +--------------------------------------+------------------+ 4 rows in set (0.04 sec) mysql:research@analytics-store.eqiad.wmnet [(none)]> SELECT table_name, (DATA_LENGTH + INDEX_LENGTH)/1024/1024/1024 as `TOTAL SIZE (GB)` FROM information_schema.tables WHERE table_name LIKE 'MobileWikiAppToC%'; +--------------------------------------+------------------+ | table_name | TOTAL SIZE (GB) | +--------------------------------------+------------------+ | MobileWikiAppToCInteraction_10375484 | 133.469544879161 | | MobileWikiAppToCInteraction_11014396 | 0.876318826340 | | MobileWikiAppToCInteraction_14585319 | 0.465894431807 | | MobileWikiAppToCInteraction_8461467 | 81.233877775259 | +--------------------------------------+------------------+ 4 rows in set (0.37 sec)
I already discussed such issues briefly with @madhuvishy and @Nuria on #wikimedia-analytics two weeks ago; back then I was informed that (despite the "slave" in the name) s1 might be more reliable than on analytics-store.
(BTW, is the information here still current? https://wikitech.wikimedia.org/wiki/MariaDB#Analytics_Slaves )