Apache Log Format Bug about lua_sandbox

[andrewhampton]

After talking a bit more in IRC to RaFromBRC, it seems it should probably actually be changed to double + integer since hitting an even number is shown as "1", not "1.000".

[trink]

I took a look at the Apache 2.4.12 code mod_log_config.c line 771

    return apr_psprintf(r->pool, "%" APR_TIME_T_FMT, apr_time_sec(duration));

Depending on what version of the APR is being used APR_TIME_T_FMT is either 'lld' or 'ld' so definitely not a decimal output. What web server/version are you using?

[andrewhampton]

I'm using hipache. https://github.com/hipache/hipache

[trink]

Hmm. not sure we want to change the Apache parser to comply with a non Apache log.

[andrewhampton]

That makes sense. Any suggestions on the best way for me to handle this?

[rafrombrc]

@trink The proposed change increases Heka's usefulness, IMO. Are there any drawbacks? Does introducing the 'or' clause hurt parsing performance?

[trink]

When parsing a real Apache log the two branches would always have to be evaluated (so technically yes it is less efficient). However, my real concern is this won't be the only difference in output (from whatever 'Apache like' source) and as time goes by we end up with a shitty grammar.

As for how do we support variants without having the user patch or replicate the grammar; I have only one thought and I am not loving it. We could expose the apache_format_variables table then people could override various specifiers from their sandbox without having to reimplement the world.

[trink]

Closing, if we open up access to the apache grammar internal variables the argument could be made to open up access for all grammars. If more use cases arrive in the future to use/modify grammar fragments this can be re-evaluated.