[Tutor] Performance Issue

Thu Oct 18 07:08:42 EDT 2018

Thanks to all for your indulgence and help…

From: Alan Gauld <alan.gauld at yahoo.co.uk> 
Sent: Thursday, October 18, 2018 6:53 AM
To: User2002 <user2002 at comcast.net>
Cc: tutor at python.org
Subject: RE: [Tutor] Performance Issue

Cc'ing list. Please use reply all on responses to tutor.

If you have no control over the server, eh access to logs etc, then the best you can do it record the time just before sending the request and immediately you get the reply. That part is outside your control. If the remaining time is worth optimising then look to your code.

As to the server time, it should be in the http headers so you don't need to parse the html, just read the headers. Much faster.

HRH,

Alan g.

On 18 Oct 2018 10:40 am, User2002 <user2002 at comcast.net <mailto:user2002 at comcast.net> > wrote:

Thank you for your thoughtful reply. There are some good ideas in there for 
me. 

I have asked and there either is no API or they do not want outsiders to 
have access to it, so I think that is a dead end. 

For my purposes, improved performance focuses on significantly reducing the 
time required to successfully execute the 
br.find_element_by_link_text(str(day_to_book)).click() command. If my 
overall time to successfully book a reservation is 1.8 seconds and roughly 2 
seconds is spent on this single instruction, then a .5 second improvement 
represents a 25% reduction. I have a second command (same type, just on the 
next iframe) that is similarly slow. So fixing both could represent a 50% 
reduction. Given the demand for the reservations (there are literally 
hundreds of people out there pounding their keyboard/clicking on fields) 
every second counts. (With an automated ability to book a reservation, I am 
probably faster than anyone's ability to click on a field, wait for a reply, 
reposition the cursor, click again, etc., but I am the point where this has 
become something I am fully invested in and would like to take as far as I 
can.) 

Most of your ideas center around the notion of knowing more about where the 
time delay occurs in the processing steps that occur outside on my world - 
communication back and forth to the server, etc. I must confess, I have no 
idea of how to do this. How can I measure what goes on outside my machine 
and measure the component parts? If you have an idea in this area or could 
refer me to where I could go to read and learn, I'd be very grateful. 

Finally, regarding your notion of web scraping, server clock, etc. Literally 
the only thing I 'scrape' is the server time to ensure I click on the date 
field at exactly 7:00:00. Once I get to that point, I click on a date field, 
then I click on a time field and I am done - no scraping occurs once it 
reaches 7:00:00. So I am not sure there are improvements to be made in that 
area. 

-----Original Message----- 
From: Tutor <tutor-bounces+user2002=comcast.net at python.org <mailto:tutor-bounces+user2002=comcast.net at python.org> > On Behalf Of 
Alan Gauld via Tutor 
Sent: Wednesday, October 17, 2018 8:12 PM 
To: tutor at python.org <mailto:tutor at python.org>  
Subject: Re: [Tutor] Performance Issue 

On 17/10/18 22:25, Stephen Smith wrote: 
> I have written a screen scraping program that watches a clock (on the 
> app's 
> server) and at 7:00:00 AM dashes to make a reservation on line. It 
> works fine. However, i have spent time trying to improve its 
> performance. I am using selenium, with chrome driver. 

When doing performance tuning the first thing to answer is what does 
improved performance mean. For example in a Word Processor improving the 
speed that an input character appears on screen by 10% is unlikely to be a 
worthwhile exercise. But improving the time taken to do a global 
search/replace by 10% might well be worthwhile. 

So what do you want to improve about an app that spends most of its time 
waiting for a change on a remote server (presumably by polling?) Is it the 
speed/frequency of polling? The speed of reading the response? The speed of 
processing the response? 

And knowing what you want to improve have you measured it to see where the 
time is being spent? Is it in the client request? The transmission to the 
server? the server processing? the transmission from the server? the reading 
of that response? or the processing of that response? You need to time each 
of those phases accurately to find out which bits are worth improving. 

> Here is what i have learned. I have tried various methods to find (by 
> link_text, by_xpath, etc.) and click on the element in question (shown 
> below). When i find the element with no click, the find process takes 
> about 
> .02 seconds. When i find it with a click (i need to select the element 
> and move to the next iframe) it takes over a second. I get these same 
> results no matter which find_element_by variation i use and i get the 
> same times in headless or normal mode. 
> 
> Here is my theory - finding the element is relatively simple in the 
> html already loaded into my machine - hence .02 seconds. However, when 
> i click on the element, processing goes out to the server which does 
> some stuff and i get a new iframe displayed, all of which takes time. 

Absolutely. network access is likely to be measured in 10ths of a second 
rather than hundredths. And processing the request may well entail a server 
database call (which may itself be on a separate machine from the web server 
with a corresponding LAN message delay), then there's the creation and 
transmission of the HTML (unless your server provides an API with JSON 
responses - but then you don't need clicks etc!) And iFrames make that worse 
since every iframe effectively gets treated as a separate html document. 

Then when your client receives the data it has to reparse the html into a 
document structure before performing the search. 

> concluded that perhaps I can't take a big chunk of that time out 

You probably can, but only if you have access to the server code and the 
network infrastructure and deep enough pockets for a server upgrade or a new 
proxy server. Assuming that's not the case then no, you need to look at 
other options. 

But your first step has to be to measure the various stages of the request. 
If the problem lies in the transmission time across the network there is 
probably not much you can do. If its in the database access (trickier to 
measure if you don't have the server code - you need to create some 
simultaneous equations using multiple test scenarios) then you might be able 
to construct better queries (eg look at a different page or only query the 
target iframe). 

> considered something other than selenium, but since i think the 
> problem lies on the server side, not sure it is worth the time. 

It depends on the nature of the page. The best solution, by far, is not to 
do web scraping. Its always the worst case solution and to be avoided if at 
all possible. Try to find an API with JSON or XML responses. 

Also, are you sure you need to use the clock on the page? 
Isn't the server clock adequate? In which case the response time should be 
in every message header so there's no need for web scraping at all... 

Finally, I think there is an active Selenium discussion forum so you could 
try there for more ideas. 

-- 
Alan G 
Author of the Learn to Program web site 
http://www.alan-g.me.uk/ 
http://www.amazon.com/author/alan_gauld 
Follow my photo-blog on Flickr at: 
http://www.flickr.com/photos/alangauldphotos 

_______________________________________________ 
Tutor maillist  -  Tutor at python.org <mailto:Tutor at python.org>  
To unsubscribe or change subscription options: 
https://mail.python.org/mailman/listinfo/tutor 