You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[FEATURE] Non-zero exit code from stackql exec when a statement fails, messages on stdout #801
stackql exec exits 0 whether or not the statement worked, and success messages go to stderr, so a script cannot tell success from failure without parsing text. Two changes would make exec usable from scripts, CI and wrappers:
Exit non-zero when a statement fails (unknown provider, service or table, SQL error, HTTP error from the provider, failed REFRESH).
Send informational messages (DDL Execution Completed, insert into table completed, github provider, version '...' successfully installed, refresh materialized view completed) to stdout, keeping stderr for errors.
A smaller related point: with --output json an empty result set prints null. [] would let consumers treat the output as an array without a special case.
Example(s)
All of these exit 0 with stackql v0.12.718:
exit=0 stdout=[] stderr=[DDL Execution Completed] <- CREATE TABLE t2 (a text)
exit=0 stdout=[] stderr=[insert into table completed] <- INSERT INTO t2 (a) VALUES ('x')
exit=0 stdout=[null] stderr=[] <- SELECT a FROM t2 WHERE a = 'none'
exit=0 stdout=[] stderr=[could not locate table 'nope'] <- SELECT a FROM nope
exit=0 stdout=[] stderr=[cannot resolve service with key = 'nope'] <- SELECT x FROM github.nope.nope
exit=0 stdout=[] stderr=[http response status code: 401, ...] <- REFRESH MATERIALIZED VIEW me (bad token)
The rule a wrapper has to use today is "stderr with nothing on stdout means failure", which is wrong for every DDL statement and for REGISTRY PULL, so those need a confirming read afterwards.
Possible Approaches or Libraries to Consider
Return an error from the exec command path when any statement in the input fails, and let cobra map it to exit code 1. Route non error messages through the normal output writer.
Additional context
Per repo HTTP failures inside a fan out (for example a 404 for one repo in a join over many repos) are arguably partial success and could keep exit 0 with the rows that did succeed, but the whole statement failing should not.
Activity