I think I've found the code location of the problem: the regex at quarry/web/app.py#241 is too restrictive. There are also database names with numbers, so simply adding an underscore won't be enough.

Also, is there any harm in trying to run a query on a validly-named but non-existing database? Users currently get a "Can't connect" message. If no harm, then I propose allowing any (semi-)valid hostname label to prevent future problems, so something like: r"^[\-0-9_a-z]+$".

The current regex excludes these dbnames from all.dblist:

arbcom_cswiki
arbcom_dewiki
arbcom_enwiki
arbcom_fiwiki
arbcom_nlwiki
arbcom_ruwiki
bat_smgwiki
be_x_oldwiki
cbk_zamwiki
fiu_vrowiki
id_internalwikimedia
map_bmswiki
nds_nlwiki
noboard_chapterswikimedia
otrs_wikiwiki
pa_uswikimedia
roa_rupwiki
roa_rupwiktionary
roa_tarawiki
sysop_itwiki
test2wiki
wg_enwiki
zh_classicalwiki
zh_min_nanwiki
zh_min_nanwikibooks
zh_min_nanwikiquote
zh_min_nanwikisource
zh_min_nanwiktionary
zh_yuewiki

Adding _ to the allowed prefix characters reduces this list to:

test2wiki

All current contents of all.dblist are matched with:

r"^(?:(?:centralauth|meta|[0-9a-z_]*wik[a-z]+)(?:_p)?)|quarry?$"

Change 676846 had a related patch set uploaded (by BryanDavis; author: Bryan Davis):

[analytics/quarry/web@master] Expand dbname validation regex

https://gerrit.wikimedia.org/r/676846

Change 676846 merged by jenkins-bot:

[analytics/quarry/web@master] Expand dbname validation regex

https://gerrit.wikimedia.org/r/676846

Mentioned in SAL (#wikimedia-cloud) [2021-04-07T21:06:15Z] <bstorm> deploying regex fixes T278715