Eh, this is pretty weak. Scrapers are no different from other browsing devices. The web speaks HTTP. There's no reason that using another HTTP browser would cause any disparate impact just by virtue of not being a conventional desktop browser -- you've thrown out a pretty absurd hypothetical. In fact, scrapers usually cause less impact because they usually don't download images or execute JavaScript.
I did an analysis and a session browsed with my specialized browser would always consume less than 100K of bandwidth (and often far less), whereas a session browsed with a conventional desktop browser would consume at least 1.2 MB, even if everything was cached, and sometimes up to 5 MB. In addition, on the desktop, a JavaScript heartbeat was sent back every few seconds, so all of that data was conserved too.
Because we were a specialized browser used by people looking for a very specific piece of data, we could employ caching mechanisms that meant that each person could get their request fulfilled without having to hit the data source's servers. We also had a regular pacing algorithm that meant our users were contacting the site way less than they would've been if they were using a conventional desktop browser.
Our service saved the data source a large amount of resource cost. When we were shut down, their site struggled for about two weeks to return to stability. I think they had anticipated the opposite effect.
Our service also saved our users a large amount of time. We were accessing publicly-available factual data that was not copyrightable (but only available from this one source's site). There's no reason that the user should be able to choose between Firefox and Chrome but not a task-specialized browser.
It is true that some people will (usually accidentally) cause a DDoS with scrapers because the target site is not properly configured, but the same thing could be done with desktop browsers. It doesn't mean that scrapers should be disadvantaged.
A small counterpoint to this -- in the airline industry, it's relatively commonplace for seat reservations to be made for a user _before_ payment has occurred. In this case, if you're mirroring normal browser activity, you can (temporarily) reduce availability on a flight, potentially even bumping up the price for other, legitimate users, and almost certainly causing the airline to incur costs beyond normal bandwidth and server costs. I'm sure there are many other domains for which this is also the case, however rare.
Cinemas and some online shops do the same. I've always wondered if it's possible to block any tickets sales for entire flights/screenings/products this way.
And if airline tickets are based on supply v demand, it might even be possible to drive down ticket prices by suddenly dropping a load of blocks near to the flight date.
Not even remotely absurd. Where is the data your scraper consuming coming from? It's almost always served from some sort of data repository (SQL or otherwise). That data costs far more per MB to serve up quickly than JS/CSS/images.
Suppose, for example, you host a blogging platform that has one very popular user. Most accounts on your site don't get a ton of visitors, and that one very popular user's post are all stored in cache.
Then along comes a scraper. He thinks, "Hey, this site is serving up a million page impressions a day. It can definitely handle me scraping the site".
But when he runs the scraper, he fills up the cache with a ton of data that it doesn't need, causing cache evictions and general performance degradation for everyone else.
There are already 6-8 major scrapers that do this constantly, across the whole internet, called search engines. You can't handle that?
What if you get a normal user who says "Hey, I wanna see some of the lesser known authors on this platform" and opens up a hundred tabs with rarely-read blogs? What if you get 10 users who decide to do that on the same day? Is it reasonable to sue them? Should there be a legal protection to punish them for making your site slow?
Don't blame the user for your scaling issues. If the optimized browser ("scraper") isn't hammering your site at a massively unnatural interval, it's clean. And if it is, you should have server-side controls that prevent one client from asking for too much data.
These are just normal problems that are part of being on the web. It's not fair to pin it on non-malicious users, even if they're not using a conventional desktop browser.
First, search engines are scrapers. No need to make a distinction.
Second, search engines don't always respect robots.txt. They sometimes do. Even Google itself says it may still contact a page that has disallowed it. [0]
Third, robots.txt is just a convention. There's no reason to assume it has any binding authority. Users should be able to access public HTTP resources with any non-disruptive HTTP client, regardless of the end server's opinion.
[0] "You should not use robots.txt as a means to hide your web pages from Google Search results. This is because other pages might point to your page, and your page could get indexed that way, avoiding the robots.txt file." / http://archive.is/A5zh8
In the Google quote you link to, Google is not contacting your page. Rather, Google will index pages that are only linked to, which it has never crawled, and will serve up those pages if the link text matches your query. That's how you get those search results where the snippet is "A description of this page has been blocked by robots.txt" or similar.
There's a somewhat related issue where to ensure your site never exists in Google, you actually need to allow it to be crawled, because the standard for that is a "<meta name=noindex ...>" tag, and in order to see the meta noindex, the search engine has to fetch the page.
And the original point of my comment was that doing this is extremely rude and not appropriate, not that it couldn't be done or that others weren't doing it.
Feel free to send any request to any server you want, it is certainly up to them to decide whether or not to serve it, but that doesnt absolve you of guilt from scraping someone's site when they explicitly ask you not to.
Please don't conflate "extremely rude", "not appropriate", and "guilt". Two of these are subjective opinions about what constitutes good citizenship. The last one is a legal determination that has the potential to deprive an individual of both his money and liberty. We're discussing whether these behaviors should be legal, not whether they are necessarily polite.
You are posting in a comment thread underneath my reply about rudeness and impoliteness, ironically being somewhat rude telling me off about what not to conflate when it was never what I said.
We do, and we also use our own user-agent string: "SiteTruth.com site rating system". A growing number of sites reject connections based on USER-AGENT string. Try "redfin.com", for example. (We list those as "blocked"). Some sites won't let us read the "robots.txt" file. In some cases, the site's USER-AGENT test forbids things the "robots.txt" allows.
Another issue is finding the site's preferred home page. We look at "example.com" and "www.example.com", both with HTTP and HTTPS, trying to find the entry point.
This just looks for redirects; it doesn't even read the content. Some sites have redirects from one of those four options to another one. In some cases, the less favored entry point has a "disallow all" robots.txt file. In some cases, the robots.txt file itself is redirected. This is like having doors with various combinations of "Keep Out" and "Please use other door" signs. In that phase, we ignore "robots.txt" but don't read any content beyond the HTTP header.
Some sites treat the four reads to find the home page as a denial of service attack and refuse connections for about a minute.
Then there's Wix. Wix sometimes serves a completely different page if it thinks you're a bot.
> I did an analysis and a session browsed with my specialized browser would always consume less than 100K of bandwidth (and often far less), whereas a session browsed with a conventional desktop browser would consume at least 1.2 MB, even if everything was cached, and sometimes up to 5 MB. In addition, on the desktop, a JavaScript heartbeat was sent back every few seconds, so all of that data was saved too.
Bandwidth is certainly part of it, but there's also also database and app-server load (which may be the actual bottleneck) that a scraper isn't necessarily bypassing.
Yeah, I just have a hard time buying that a scraper that does less than a conventional desktop browser is going to accidentally stumble across something that causes the server-side to flip out. I'm not really sure in what case your hypothetical is plausible.
Scrapers are usually used to get publicly-available data more efficiently. What you're describing would basically require the scraper to hammer an invisible endpoint somewhere, but there's no reason the scraper would do that -- it just wants to get the data displayed by the site in a more efficient manner. I suppose the browser could enforce a cooldown on an expensive callback via JavaScript, which a scraper would circumvent, but IMO that's not a fair reason to say scrapers are disallowed; cooldowns should be enforced server-side. There's no way to ensure that a user is going to execute your script. That's just part of the deal.
Everything about scrapers means less server load; no images, no wandering around the site trying to find the right place, no heavy JavaScript callbacks that invoke server-side application load, etc. Scrapers are just highly-optimized browsing devices targeting specific pieces of data; it's logical that they would be cheaper to serve than a desktop user who's concerned about aesthetics and the like.
In our specific case, those JavaScripts we didn't download included instructions to make over 100 AJAX requests on every page load. No wonder users were looking for something more efficient.
So I agree that a scraper isn't necessarily bypassing some load-heavy operations, but I find it highly implausible that a non-malicious scraper would be invoking operations that cause extra load (beyond just hitting the site too often). Frankly, I'd be surprised if there was a functional scraper that regularly invoked more resource cost per-session than a typical desktop browsing session to get equivalent data.
> What you're describing would basically require the scraper to hammer an invisible endpoint somewhere
That wasn't my point. My point was: a lot of a website's costs are hidden from a web scraper (e.g. database load), so a scraper can't claim, based on the variables they can observe (bandwidth), that they're costing the website less than normal traffic.
I was basically responding to statements like this:
> In fact, scrapers usually cause less impact because they usually don't download images or execute JavaScript.
There's really no way for a scraper to know that unless the website tells them. Their usage pattern is different than typical users and raw bandwidth (for stuff like static images) may not matter to the website.
It's true that there's no way to know that for sure, but it doesn't make sense that a scraper, by virtue of its being a scraper, is incurring additional load. A scraper is only making requests that a person with a desktop browser or any other appliance that speaks HTTP could make. What's the difference between a user clicking the same button on the page 50 times or holding down F5 and a scraper that pings a page once a minute?
Your argument is basically boiling down to "scrapers could hit one load-heavy endpoint too fast", but so could desktop browsers. So I don't see what it has to do with scraping.
> but it doesn't make sense that a scraper, by virtue of its being a scraper, is incurring additional load
It does, because scrapers don't have normal usage patterns. They're robots and behave like robots.
> What's the difference between a user clicking the same button on the page 50 times or holding down F5 and a scraper that pings a page once a minute?
Typical users aren't usually in the habit of mashing F5, especially not for robotically long periods of time. It's basically the difference between a theoretical activity and an actual activity.
Basically, scraping is not regular usage, and I don't think it's correct to pretend that they're equivalent (or more extremely, that scraping is less costly to the website).
Scrapers are usually coded to have as regular of a usage pattern as possible, so that the data they retrieve is as much like the data the end user would receive as possible.
For example, Googlebot does everything in its power to ensure that it sees pages the same way that end users sees them, executing JavaScript and performing OCR to try to read information conveyed in images. Google also has non-Googlebot scans to try to determine if a page is serving different content to Googlebot-labeled scans, and they penalize sites that they suspect of doing this.
While it is true that someone could write a scraper that obviously behaved robotically, it is also true that someone could use their desktop browser in a robotic way. Mashing F5 is so common that there are many ancient memes referring to and making jokes about that activity. There are extensions that end users use to record browser macros, behaviors they want their browser to repeat over and over again.
However, this conversation about whether scrapers behave robotically or not is moot because a web site shouldn't break down under load when someone uses it in a slightly-irregular way. The obvious, crappy scrapers are trivial to block. The ones that blend into the traffic are no harm, no foul. If you can't tell the difference between an optimized browser like a scraper and a general-purpose browser like Chrome, why shouldn't it be allowed to talk to your site?
> Typical users aren't usually in the habit of mashing F5, especially not for robotically long periods of time. It's basically the difference between a theoretical activity and an actual activity.
Just like every university site ever is completely down during signup days because everyone is mashing F5.
Link me your site, I’ll treat it like a college student waiting to be able to sign up for their classes.
I did an analysis and a session browsed with my specialized browser would always consume less than 100K of bandwidth (and often far less), whereas a session browsed with a conventional desktop browser would consume at least 1.2 MB, even if everything was cached, and sometimes up to 5 MB. In addition, on the desktop, a JavaScript heartbeat was sent back every few seconds, so all of that data was conserved too.
Because we were a specialized browser used by people looking for a very specific piece of data, we could employ caching mechanisms that meant that each person could get their request fulfilled without having to hit the data source's servers. We also had a regular pacing algorithm that meant our users were contacting the site way less than they would've been if they were using a conventional desktop browser.
Our service saved the data source a large amount of resource cost. When we were shut down, their site struggled for about two weeks to return to stability. I think they had anticipated the opposite effect.
Our service also saved our users a large amount of time. We were accessing publicly-available factual data that was not copyrightable (but only available from this one source's site). There's no reason that the user should be able to choose between Firefox and Chrome but not a task-specialized browser.
It is true that some people will (usually accidentally) cause a DDoS with scrapers because the target site is not properly configured, but the same thing could be done with desktop browsers. It doesn't mean that scrapers should be disadvantaged.