#haraka

_smf_: I've updated the PR :)

This is more of a request but could I get a patch release when the logfmt is merged? Else I'm going to have to pull from the git repo when doing `npm i -g Haraka`. If no, what needs to be done? I can try and help.

anosh: Well I'm not doing a patch release today - I don't have time. npm can install packages directly from Git anyway.

Yeah

I'm fine with that

:)

And I'm not happy with the PR as it is.

I'm now about to benchmark those logging changes as I didn't appreciate exactly what you had done.

If they turn out to be slower (and I highly suspect they will), then I'm going to be requesting further changes.

Alright, no probs

anosh: I'd seriously be considering if logfmt is actually worth the penalty if I were you.

I'm running benchmarks now betweek old logger, new default, new logfmt

I've already done it.

See the pull request

just saw

On a really basic test - it over 160% slower.

Which TBH is hardly surprising given all the dependencies it has.

If you benchark it inside Haraka then I suspect the results will be worse.

I didn't do that as it's time consuming and ultimately it won't make any difference to my views on it.

I've been getting a 2 sec difference :/ with Haraka https://gist.github.com/KingNoosh/5e2d4c85a01ce...

I understand that template literals are faster, the issue would be somehow knowing there are key=value already to build the end result in logfmt instead

aaaah let me see what I can do

anosh: how did you do that test? What loglevel etc.

protocol

I ran `time smtp-source -m 8000 127.0.0.1:2500`

with ok-ing both rcpt and queue

Ok - so we tested different things.

yeah

My answer is still the same.

There is no way I want logfmt in Haraka by default.

Alright

I can gist you the code I used if you want to try it for yourself.

That'd be great thanks

thanks :)

hey

once again i wanna to ask about tls config - i got error (No Valid TLS config) - i see that lines added in tls.js 6 days ago. I create bundle with key/cert/cat in configt/tls/domain_name.pem. Still got 'no valid tls config'. Where now should be places all tls confs?

gruceqq: you said you were using 2.8.14 yesterday. So what you actually meant is that you installed Haraka from git then?

ultimatt made a bunch of TLS changes a few days ago. See https://github.com/haraka/Haraka/commit/1e20063...

If it doesn't work for you, then raise an Issue on github.

gruceqq: did you definitely install Haraka from github?

As looking at the code, this would seem like a rather large bug.

_smf_ after last problems i do everything from scratch - clone from githum, npm install (dependencies)

gruceqq: raise an issue then please.

You might want to actually install the release version of Haraka instead.

e.g. npm install -g Haraka

I know I can rid of logfmt, I don't care about anything else besides the stringify function, but I just need to make that fast

the module*

i revert to relase version and everything works fine. port 2555 for smtp_forward/proxy should be added to smtp.ini listen ports?

gruceqq: only if you want to create a mail loop.

gruceqq: you use smtp_forward or smtp_proxy plugins to forward mail to another mail server.

:D i didn't wanna to create mail loop _smf_ :D tryin to get rid off from this - [core] Plugin queue/smtp_proxy failed: Error: This socket is closed

sure _smf_

I think I've actually made it faster

than the old log

this is so exciting

morning

morning o/

morning baudehlo

morning :)

_smf_: I'm getting around the same with logger1 and this: https://gist.github.com/KingNoosh/6188838359fc3...

that's the improved one compared to the stupid node module

Could you run it to see whether it's good for you whenever you get the time?

anosh: I can't believe that is going to be anywhere near the performance given that it uses regexps.

I'll give it a try though.

As I expected:

smf@i7desktop:~$ time node haraka_logger5.js 2>&1 > /dev/null

real0m7.255s

user0m7.080s

sys0m0.244s

67.17% slower

For 1 million iterations

yeah

It was using indexOf before

switching to regex made it faster for me

fuck

This probably varies on node versions fwiw.

I'm on 6

idk how _smf_ is getting a large gap too which is frustrating

I'm on 7.10.1 on this machine.

Node 8 includes an entirely different JS engine.

baudehlo: I know, but whilst the performance might alter a bit. Running regexp over every single key is always going to be slower than string concatenation isn't it.

Not necessarily.

Ok - but do you want to modify logger and make it slower than it already is for a feature that so far one person is submitting a PR for?

regexp was running faster than 4 indexOf's

Can I see the comparison code?

The original pull was >160% slower than our stock code for 1 million operations.

baudehlo: here you go: https://gist.github.com/smfreegard/a6c859fa6b5b...

.vs.

Note that there's a bug in anosh's code there. It's doing an if (!foo) .vs. if (foo !== undefined), so it fails on numeric 0.

It was originally a `foo !== undefined || foo !== null`

At this point I'm trying to just make it faster

anosh: if you run it as it is though, errors= is reported incorrectly.

So it's got to be fast and correct.

For a fair test, shouldn't `var logdetail` be constructed in the for loop?

yeah

is $& expensive in node like it is in Perl?

Like maybe use () and $1 instead.

baudehlo: if I move the logdetail into the loops, it's 54.6% slower .vs. 67.17% when outside of the loop.

I constructed it outside of the loop to compare the functions that were building the log line, but I can see that creating an Array .vs. Object would probably have an effect too.

I 'll try that baudehlo

baudehlo: I mean arguably a 50% time increase probably isn't going to be particularly noticable overall given this is rather artificial.

The `if` is correct actually because it does a toString() and "0" is true in node.

OK, this version of stringify is optimized now (V8 didn't like a few things about the other version):

Still a lot slower, but it's progress.

baudehlo: your version is actually slower for me.

Yeah but at least the compiler optimizes it.

baudehlo: actually _smf_, 0 isn't truthy, I did `let value = (!obj[key] && obj[key] !== 0) ? obj[key] : '';`

node doesn't consider it truthy

I double checked

_smf_ was right*

anosh: no - 0 is false but "0" is true

e.g. a string containing zero.

OOH

Yeah

I misunderstood then