asciilifeform: lobbes: i'ma try your proggy tomorrow. ideally time finally came to eat all of the back-archives.

asciilifeform: lobbes: possibly it can even be used to bake a phf-style refill/sync system that eats znc.

asciilifeform: lobbes: http://blog.lobbesblog.com/2019/08/znc2tmsr_etckvvpatch/comment-page-1/#comment-59

asciilifeform: diana_alt: there defo is. i've measured up to 0.6s (and all outside of the py proggy per se, i.e. vanishes if accessed at localhost )

asciilifeform: diana_alt: fwiw i was not, to date, able to rule out mp's 'overworked pipe' hypothesis .

asciilifeform: when someone sets up this logger outside of piz, i suppose will tell us sumthing re subj.

asciilifeform: diana_alt: i do not currently know, atm i only have access to 1 piz machine. if you have time, plz try from yours and share result.

asciilifeform: diana_alt: however, if accessed from localhost on ~public port~ (80) then yes same delay

asciilifeform: so i suspect you will find same

asciilifeform: delay disappears when accessed on localhost at the 'high' (nonpublic) port where the py proggy itself is answering

asciilifeform: pipe delay ( piz <-> asciilifeform's chair , cannot be generalized to entire planet ) varies from 0.2-0.4s (for revvup of tcp pipe, that is)

asciilifeform: diana_alt: if you press the vtree and set up the logger on own box, where you happen to be, and import the daily db dump, can then observe the 'native' speed of the thing.

asciilifeform: i get (and on rather slow machine) load time for the 'heavy' example page comparable to the time it takes for the mouse button to go up an' down.

asciilifeform: diana_alt: i have plan to move the box to traditional apache, but yet not figured out how to do so w/out unacceptably long disruption of service .

asciilifeform: ( at one pt, long ago, when baked phuctor, asciilifeform had a preference for nginx, but it turned out to be ill-founded . currently both phuctor and log www are served by a nginx . )

asciilifeform: neither phuctor nor logger is written in such a way as to force the choice of either nginx or apache as the front-side proxy, will eat either just the same. simply requires the appropriate config.

asciilifeform: the reason why i currently suspect nginx as the culprit, is that i observed very similar delay when loading static (e.g. FG photograph) files via same.

asciilifeform: previously i chalked this up to 'slow pipe'. now suspect also the proxying proggy.

asciilifeform: ( had not, prior to folx observing 'logger is slow, say wai' , taken the sweat to properly monitor where in the typical page load is time spent and on what )

asciilifeform: BingoBoingo: plz perform the simple experiment suggested by diana_alt and tell result.

asciilifeform: diana_alt: reasons to suspect the pipe, include this: i have stood up temporarily a raw flask on port 33. try time curl "http://logs.nosuchlabs.com:33/log" > /dev/null

asciilifeform: ( diana_alt lemme know after tried, i'ma switch it off, it's an unhygienic thing )

asciilifeform: in my own test, there is no substantial detectable diff

asciilifeform: ( keep in mind that the links in the loaded page still point to the original )

asciilifeform: diana_alt: do 3-4 shots via each (main vs 33) and paste plox

asciilifeform: ty diana_alt , BingoBoingo . (i'ma switch off the 2nd logger for nao, will run it again if anyone requests, later)

asciilifeform: BingoBoingo's output is pretty interesting -- how is it that he gets >2x the wait from ~local~ machine than diana_alt from europistan.

asciilifeform: ( possibly BingoBoingo's is under load ?? )

asciilifeform: diana_alt's, on other hand, is very similar to mine, i.e. no substantial diff b/w main and 'naked'

asciilifeform: BingoBoingo: naked ip also worx (lemme know if you want the 2nd logger on 33 switched on , to try moar )

asciilifeform: the logger does not depend on dnsism to work ( can set the domain in /etc/hosts like e.g. mp does )

asciilifeform: BingoBoingo: the 33 logger is off atm

asciilifeform: BingoBoingo: that loox realistic.

asciilifeform: ( and imho further suggests that pipe is culprit )

asciilifeform: 'go to war with the pipe you have, not the pipe you wish you had'(tm)(r)

asciilifeform: piz pipe is not gonna set any speed records. (however, is extremely reliable, e.g. asciilifeform gets ssh sessions that last 3-4 months)

asciilifeform: ... and typically they die on acct of connectivity pause on asciilifeform's end, rather than piz

asciilifeform: !quptime

asciilifeform: ^ the log bot itself, for instance, thus far has only reconnected when asciilifeform manually cycled it, to update to new ver (last such time, to put phf's uniturd pill in production)

asciilifeform: interesting.

asciilifeform: oook here's an experiment : asciilifeform set up a mirror of the logotron on a cheap heathen host. ( one of the most rubbish hosters known, lulazon's . hence cannot in fact stand up copy of bot there, fleanode bans conns from it unless sslized. but the db is current. )

asciilifeform: time curl http://54.209.217.34/log > /dev/null from my chair yields 0.063s.

asciilifeform: other folx welcome to try. e.g. the 'heavy page' : curl http://54.209.217.34/log/trilema/2018-04-18#1802801 > /dev/null

asciilifeform: ... from here eats 0.268s.

asciilifeform: deliberately this setup uses nginx and same (as far as could replicate on the heathen box) stack of proggies as the orig.

asciilifeform: this suggests that culprit of 'mystery delay' is in fact piz pipe.

asciilifeform: errybody's welcome to play with the mirror ( it's a '1st month -- phreeee!' pseudo-machine , i'ma leave it up for a spell, tho it will , obv., grow no new log lines )

asciilifeform: i suppose for completeness will mention an alt-hypothesis -- somehow nic on dulap is culprit.

asciilifeform did not bother to sync the clock, so times on shit mirror will vary vs. dulap's

asciilifeform: all knobs aside from 'download snapshot' work, so feel free to e.g. walk the 'random' pages etc.

asciilifeform: this experiment also reveals that asciilifeform's logotron will in fact stand up, with no modifications, on a python, flask, pg, etc. plucked out of arbitrary rubbish bin.

asciilifeform: i'ma switch this thing off in coupla days, once errybody's taken the offered measurement and is satisfied ( i dun want a 'fossilized' logger sitting for idiots to find and think 'they're dead' )

asciilifeform: BingoBoingo, diana_coman, et al ^ as time permits , plox to post load times of this ^ log.

asciilifeform brb,teatime

asciilifeform: BingoBoingo: hard to say for fact, but seems to add up to a picture where's there's a ~0.5s bottleneck b/w south amer. and errywhere else.

asciilifeform: ( in the mirror example, we get it in the reverse direction, but seems to be that same ~0.5s ! )

asciilifeform: other folx btw are encouraged to perform similar experiment ; whole thing took me approx 35min of meat time, from revving up heathen pseudo-box to posting link here.

asciilifeform: ( of which ~8m spent downloading db snapshot from piz )

asciilifeform: ( the rest -- largely, hand-copying configs )

asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-08-19#1929694 << this is worth expanding on. asciilifeform also gets very fast pings ( never moar than 200msec, to date ) and fast udp. what i suspect is, usg's snoop gear that sits as parasitic toad on south amer's pipe, specifically slows down tcp.