Show HN: WarcDB: Web crawl data as SQLite databases

https://github.com/Florents-Tselai/WarcDB

by fforflo • 4 years ago

171 30 4 years ago

uniqueuid6• 4 years ago

Since this is pretty new, some background:

WARC is a file format written and read by a rather small but specialized set of web crawler tools, most notably the internet archive's tooling. For example, its java-based crawler heretrix produces warc files.

There are a couple of other very cool tools, such as warcprox, which can create web archives from web browser activity by acting as a proxy, pywb which play back the same to make archived versions browsable, and related libraries [1](shoutout to Noah Levitt, Ilya Kreymer and collaborators for building all this).

The file format itself is an iso standard, and it's very simple: Capture all http headers sent and received, and all http bodies sent and received, and do simple de-duplication based on digests of the body.

There is a companion format, CDX, which builds indexes from warc files (which in turn are just concatenated records, so rather robust).

Although all of this is great, I worry a bit about where we're heading with QUIC / udp-based protocols, websockets and other very involved protocols which ultimately make archival much harder.

If there's anything you can do to help these (or other) tools to keep our web's archival records alive and flowing, please do so.

[1] https://github.com/internetarchive/warcprox

fforflo5• 4 years ago

The web archiving community is surprisingly small and fragmented (in terms of tools) given its impact. Thankfully the .warc format looks pretty powerful and standard for the web we have so far (which is a lot! ).

Now with the new protocols, dunno maybe its too soon to worry? Then again, maybe its an IPv4 / IPv6 analogy.

There is a companion format, CDX, which builds indexes from warc files (which in turn are just concatenated records, so rather robust).

Good point. I's planning of combining this fact with the ATTACH option that SQLite has - allowing to query multiple database files [0]

[0] https://www.sqlite.org/lang_attach.html

uniqueuid6• 4 years ago

Oh hi, thanks for building this! I haven't had the chance to play with it, but my hunch is that sqlite for warc can fill a great niche and would be much more portable (and probably performant).

Allowing multiple DB files is a great idea, since that fundamentally enables large archives, cold-storage and so on.

pstuart• 4 years ago

Swarm and Union should be of interest to you:

  https://www.sqlite.org/swarmvtab.html
  https://www.sqlite.org/unionvtab.html

ma2rten• 4 years ago

Common Crawl is also in WARC format.

lijogdfljk2• 4 years ago

Would a WARC format reduce effort needed to make Reader-like programs? Ie strip pages of HTML cruft, leaving you with text, images, etc - the content?

uniqueuid6• 4 years ago

Not at all.

A WARC file gives you exactly what you would see on the wire, or in the network inspector tab of your browser. It does nothing to the content, and that's the point.

The only thing you gain (and that's very important for other reasons as well) is an immutable ground truth to work from when creating the reader view of a given article.

simonw• 4 years ago

I've found the Readability.js library to be really good for that - here's my recipe for running it as a CLI: https://til.simonwillison.net/shot-scraper/readability

fforflo5• 4 years ago

I built this as a small utility within a larger project I'm working on these days. (contact me if you're curious or want to support).

The WARC format is extremely simple and yet so powerful. Most importantly though, there are already pebibytes of already crawled archives.

This is a fairly straightforward mapping of a .warc file to a .sqlite database. The goal is to make such archives SQL-able even in smaller pieces.

The schema I've come up with it's tailored around my requirements, but comment if you can spot any obvious pitfalls.

PS: I do believe that at some point .sqlite will become the defacto standard for such initiatives. Sure, it's not text... but it's pretty close.

marginalia_nu3• 4 years ago

PS: I do believe that at some point .sqlite will become the defacto standard for such initiatives. Sure, it's not text... but it's pretty close.

What is the advantage of moving around .sqlite-files, over just loading the (compressed) WARCs into sqlite databases when you need them?

I've been messing around with different formats for my own search engine crawls, and ended up with the conclusion that WARC is a pretty amazing intermediary format that weighs both the needs of the producer and consumer very well. I don't use WARCs now, instead something similar, but I probably will migrate toward that format eventually.

WARC's real selling pint is that it's such an extremely portable format.

uniqueuid6• 4 years ago

Anecdotal evidence, but I produced a medium-size crawl in the past (~20TB compressed). I used distributed resources with off-the-shelf libraries (i.e. warcprox etc.) and managed to get corrupted data in some cases where neither the length-delimited (i.e. offset + payload length) nor the newline-delimited (triple newlines between records) logics were valid any longer. Took me some time to build a repair tool for that.

Sqlite has an amazing set of well-understood and documented guarantees on top of performance, there's a host of potential validation tools to choose from and you can even use transactions etc. So that alone seems like a great idea.

What's more, you can potentially skip CDX files if you have sqlite databases (or build your own meta sqlite database for the others quickly).

fforflo5• 4 years ago

What is the advantage of moving around .sqlite-files, over just loading the (compressed) WARCs into sqlite databases when you need them?

The .warc spec is ideal. I'm not saying we replace it (ref. xkcd: standards).

On top of what uniqueid said, "loading" is much slower and more cumbersome than it sounds. I'm not saying SQLite will replace text (maybe my aphorism sounded too firm). I'm saying that maybe along with the .warc.gz archives at rest, one could have .sql.gz files at rest as well.

In other words: why not move ACID-compliant archives moving around?

mynameismon• 4 years ago

What is the advantage of moving around .sqlite-files, over just loading the (compressed) WARCs into sqlite databases when you need them?

I suppose there could be made a case for easy extension: You don't need to change the entire spec to add another table in the SQLite database, maybe containing other metadata.

WARC's real selling pint is that it's such an extremely portable format.

I mean, so is SQLite: it is also apporoved as a LoC archival method. (See SQL Archive [0])

[0] https://www.sqlite.org/sqlar.html

nlohmann• 4 years ago

Have you every played with SQLite virtual tables (https://sqlite.org/vtab.html) - they could allow to provide an SQLite interface while keeping the same structure on disk. Though it requires a bit of work (implementing the interface can be tedious), it can avoid the conversion in the first place.

fforflo5• 4 years ago

Good point. Actually CommonCrawl provides Parquet files for their archives too.

And there's this vtable for Parquet extension. https://github.com/cldellow/sqlite-parquet-vtable

But for my use case virtual would be too complicated.

fforflo5• 4 years ago

Then again, do you need virtual tables? The .warc structure won't change, so the tables won't change. But you can have SQL views defined instead for common queries.

jbverschoor• 4 years ago

I like that you're logging responses instead of just the result / payload. Reminds me of some idea I had of using something like queue as an intermediary between a webserver and the backend. I don't exactly remember my reasoning anymore right now

tepitoperrito• 4 years ago

It'd be neat to extend the warc format and tooling to support cached http responses for things like REST endpoints. Then you could make sure everything you did in a session is recorded for later use.

From the specification it would appear fairly straightforward once an approach was chosen...

Here's the relevant extract from the warc spec that informed my difficulty estimate - "The WARC (Web ARChive) file format offers a convention for concatenating multiple resource records (data objects), each consisting of a set of simple text headers and an arbitrary data block into one long file."

Edit: Upon 2 minutes of reflection I think the way to go for what I'm envisioning is some kind of browser session recording -> replayable archive solution.

uniqueuid6• 4 years ago

Although this is a nice idea, it's extremely difficult to get it completely right.

Consider a SPA where navigation happens via xhr or similar requests and updates are json that's patched into the DOM. Even browsers have a hard time figuring out how to make this a coherent session.

Now with warc, you get a single record per transfer, i.e. every json file, every image, every css is an individual record. It's completely up to the client/downstream tech to re-assemble this into a coherent page.

If you want to go down that road, my best suggestion would be to start with a browser's history - that's probably the most solid version of a session that we have right now.

nikisweeting• 4 years ago

ArchiveWeb.page + ReplayWeb.page are able to achieve what you're describing.

TedDoesntTalk• 4 years ago

Did you add this to https://github.com/dhamaniasad/WARCTools ?

Show HN: WarcDB: Web crawl data as SQLite databases

Related Stories