This module contains the gRPC schema used in a URL Frontier implementation as well as the Java code generated from it.
An automatically generated overview of the endpoints and messages can be found here
The endpoints and messages described by the API are meant to be a minimum. Specific implementations can offer additional functions on top of the ones described here.
The two main methods used by a crawler to interact with a URL Frontier service (let's just call it service from now on) are:
- GetURLs
- PutURLs
Underpinning them is the concept of queue(s) and crawl(s).
What the queues should be based on is determined by the client, through the setting of a string value (key) associated with the messages sent with the PutURLs method. The value could be the hostname of a URL, its domain, IP or anything else. An empty value leaves the service to route the messages into a queue - the hostname being the default behaviour. It is up to the client code to be consistent in the use of the keys.
The keys are used in several functions: GetStats, DeleteQueue and GetURLs.
A service can handle one or more crawls, identified by a crawlID. Queues with the same keys can exist in multiple crawls and will be treated as totally distinct instances, i.e. deleting a queue for a particular crawl will not affect the queues with the same keys in other crawls. Similarly, a URL is considered unique to a crawl.
A crawl can be thought of as a namespace. The default crawlID is 'DEFAULT' but it is expected that service implementations will handle empty strings as an equivalent (note: this is due to Protobuff not allowing to have default values for String fields).
The service returns URLs ready to be fetched. It helps enforcing politeness by limiting the number of URLs per queue to be returned as well the amount of time to wait for until the URLs returned will be eligible to be returned again. This is used to prevent URLs to be in limbo if the client code crashes and is resumed later. It is easier to think about the URLs that have been returned by the GetURLs function as being in transit. They remain so until an update is received for them via the PutURLs function (see below) or the value set in delay_requestable has elapsed.
Internally, the service will rotate on the queues to be pulled from and will aim at an optimal distribution. Multiple clients can call getURLs on a single instance and will each get URLs from across queues as URLs are prioritized by the service.
This method is called to add newly discovered URLs to the frontier (e.g. they have been found while parsing a HTML document or a sitemap file) but also to update the information about URLs that had been previously obtained from GetURLs and have then been processed by the crawler. The latter allows to remove them from the in transit status and so, more URLs can then be returned for its queue. Arbitrary metadata can be associated with a URL, for instance to store the depth of links followed since injecting the seeds or the HTTP code last obtained when a known URL has been fetched.
Discovered URLs are treated differently from known ones which are being updated. Discovered URLs will be added to the queues only if they are not already known, whereas known URLs will always be updated.
Another difference is in the scheduling of the URLs: discovered URLs are added to the queues (if they are unknown so far) without specific information about when they should be fetched - the service will return them as soon as possible. Known URLs on the other hand can have a refetchable_from_date meaning that the service will put them back in the queues and serve them through getURLs when the delay has elapsed. This is useful for instance when a transient error has occurred when fetching a URL, we might want to try it later. If no value is specified, the URL will be considered done and won't be returned by getURLs ever again.
The URLs are sorted by refetchable_from_date, which is typically the number seconds of UTC time since Unix epoch. The frontier checks that this value is lower or equal to the current timstamp in order to emit them. With that in mind, you can set any value you want as long as it is not 0 to prioritise URLs within a queue.
Some of the messages used by the API have a field local. This is used to indicate whether the action is pertaining only to the target node or to the whole cluster. For instance, the method GetStats can return either the stats for the particular instance of the Frontier or the whole cluster.
The filtering logic has to be handled within the crawlers as it is often application specific.
The robots directives are not stored within the URL Frontier.
The Java code generated from the schema is available as a Maven dependency.
<dependencies>
<dependency>
<groupId>com.github.crawler-commons</groupId>
<artifactId>urlfrontier-API</artifactId>
<version>2.3</version>
</dependency>
</dependencies>
The Java code can be (re)generated as follows; change the OS & processor values if required.
osproc=linux-x86_64
wget https://github.com/protocolbuffers/protobuf/releases/download/v3.20.3/protoc-3.20.3-$osproc.zip
unzip -p protoc-3.20.3-$osproc.zip bin/protoc > protoc
rm protoc-3.20.3-$osproc.zip
chmod a+x protoc
wget https://repo1.maven.org/maven2/io/grpc/protoc-gen-grpc-java/1.50.2/protoc-gen-grpc-java-1.50.2-$osproc.exe
chmod a+x protoc-gen-grpc-java-1.50.2-$osproc.exe
./protoc --plugin=protoc-gen-grpc-java=./protoc-gen-grpc-java-1.50.2-$osproc.exe --proto_path=. --java_out=src/main/java --grpc-java_out=src/main/java urlfrontier.proto
Since the Java code is provided here and the corresponding JARs will be available from Maven, regenerating from the schema is not necessary.
For other languages, you need to generate the code stubs yourself, as shown here for Python
python3 -m pip install grpcio-tools
mkdir python
python3 -m grpc_tools.protoc -I. --python_out=python --grpc_python_out=python urlfrontier.proto
Alternatively, docker-protoc can be used to generate the code in various languages:
docker run -v `pwd`:/defs namely/protoc-all -f urlfrontier.proto -l java -o src/main/java
docker run --rm -v $(pwd):/out -v $(pwd):/protos pseudomuto/protoc-gen-doc --doc_opt=markdown,urlfrontier.md
The current version of the documentation can be found here