Broad web crawling is an activity of going through practically boundless web by starting from a set of locations(urls) and following outgoing links
It presents a unique set of challenges one must overcome to get a stable and scalable system, Crusty is an attempt to tackle on some of those challenges to see what's out here while having fun with Rust ;)
This particular implementation could be used to quickly fetch a subset of all observable internet, discover most popular domains/links
Built on top of crusty-core which handles all low-level aspects of web crawling
Configurability && extensibility
see a typical config file with some explanations regarding available options
Fast single node performance
Crusty is written in Rust on top of green threads running on tokio, so it can achieve quite impressive single-node performance even on a moderate PC
Additional optimizations are possible to further improve this(mostly better html parsing, there are tasks that do not require full DOM parsing, this implementation does full DOM parsing mostly for the sake of extensibility and configurability)
Additionally, Crusty has small and predictable memory footprint and is usually cpu/network bound. There is no GC pressure and no war over memory.
Scalability
Each Crusty node is essentially an independent unit which we can run hundreds of in parallel(on different machines of course),
the tricky part is job delegation and domain discovery which is solved by a high performance sharded queue-like structure built on top of clickhouse(huh!).
One might think, "clickhouse? wtf?!" but this DB is so darn fast while providing rich querying capabilities, indexing, filtering.
The idea is basically a huge sharded table where each domain belongs to some shard(crc32(domain_name) % number_of_shards), now each Crusty instance can read from a unique subset of all those shards while can write to all of them(so called domain discovery).
I.e. each shard is readable by only one Crusty instance by a single thread(it's important, to avoid interference in multi-node setups).
On moderate installments(~ <16 nodes) such systems is viable as this, although if someone tries to take this to a mega-scale dynamic shard manager might be required...
There is additional challenge of domain discovery deduplication in multi-node setups, - right now we dedup locally and on clickhouse(AggregatingMergeTree) but the more nodes we add the less efficient local deduplication becomes
In big setups a dedicated dedup layer might be required, alternatively one might try to simply push overflowing deduplication jobs to clickhouse by creating more shards
Basic politeness
While we can crawl thousands of domains in parallel - we should absolutely limit concurrency on per-domain level
to avoid any stress to crawled sites, see job_reader.default_crawler_settings.concurrency.
It's also a good practice to introduce delays between visiting pages, see job_reader.default_crawler_settings.delay.
Additionally, there are massive sites with millions of sub-domains(usually blogs) such as tumblr.com.
Special care should be taken when crawling them, as such we implement a sub-domain concurrency limitation as well, see job_reader.domain_tail_top_n setting which defaults to 3 - no more than 3 sub-domains can be crawled concurrently
Currently, there's no robots.txt support but this can be added easily(and will be)
Observability
Crusty uses tracing and stores multiple metrics in clickhouse that we can observe with grafana - giving a real-time insight in crawling performance
install docker && docker-compose, follow instructions at
https://docs.docker.com/get-docker/
https://docs.docker.com/compose/install/
Crusty source code
git clone https://github.com/let4be/crusty
cd crusty
study config file and adapt to your needs,
there are sensible defaults for a 100mbit channel, if you have more/less bandwidth or poor cpu you might need to adjust concurrency_profile
build docker-compose build
run CRUSTY_SEEDS=https://example.com docker-compose up (abortable with ctrl+c)
to run in background
CRUSTY_SEEDS=https://example.com docker-compose up -d
to stop background run and retain crawling data
docker-compose down
to stop background run and erase crawling data(clickhouse/grafana)
docker-compose down -v
see Crusty live at http://localhost:3000/d/crusty-dashboard/crusty?orgId=1&refresh=5s
to see running containers docker ps(should be 3 - crusty-grafana, crusty-clickhouse and crusty)
to see logs: docker logs crusty
if you decide to build manually via cargo build, remember - release build is a lot faster(and default is debug)
In the real world usage scenario on high bandwidth channel docker might become too expensive,so it might be a good idea either to run directly or at least in network_mode = host
just use docker-compose, it's the recommended way to play with Crusty
however...
to create / clean db use this script(must be executed -in context- of clickhouse docker container)
grafana dashboard is exported as json model
I'm open to discussions/contributions, - use github issues,
pull requests are welcomed ;)