The warc->slop converter was rejecting some items because they had headers that were representable in the Warc code's MessageHeader map implementation, but illegal in the HttpHeaders' implementation.
Fixing this by manually filtering these out. Ostensibly the constructor has a filtering predicate, but this annoyingly runs too late and fails to prevent the problem.
The crawler was incorrectly using the request URL as the base URL when resolving relative links. This caused problems when encountering redirects.
For example if we fetch /log, redirecting to /log/ and find links to foo/, and bar/; these would resolve to /foo and /bar, and not /log/foo and /log/bar.
Nephentes has been doing the rounds in social media, adding an easy detection and mitigation mechanism for this type of trap, as sadly not all webmasters set up their robots.txt correctly. Out of the box crawl limits will also deal with this type of attack, but this fix is faster.
This gives the same upper limit to the live crawler and the big boy crawler, though the live crawler will reject items too large, and the big crawler will truncate at that point.
We should not wait until we've fetched robots.txt to decide whether we have any data to fetch! This makes the live crawler very slow and leads to unnecessary requests.
Discovered feed URLs were given a double slash after their domain name in the DB. This will go away in the URL normalizer, so the URLs are still viable, but the commit fixes the issue regardless.
Feed discover is improved with by probing a few likely endpoints when no feed link tag is provided. To store the feed URLs, a sqlite database is added to each crawlset that stores a simple summary of the crawl job, including any feed URLs that have been discovered.
Solves issue #135
Updated the `documentBody` method to improve parsing retries and error handling. Refactored `ContentType` charset processing with cleaner logic, removing redundant handling for unsupported charsets. Also, updated the version of the `slop` library in dependency settings.
Also alter CrawledDocument to not require String parsing of the underlying byte[] data. This should reduce the number of large memory allocations quite significantly, hopefully reducing the GC churn a bit.
Replaces Parquet output and processing with the new Slop-based format. Includes data migration functionality, updates to handling and writing of crawl data, and introduces support for SLOP in domain readers and converters.
Added an additional filter step to ensure URLs with binary suffixes are excluded during crawling. This prevents unnecessary processing of non-HTML content, improving the efficiency of the link parsing process.
Since some of the export tasks have been memory hungry, sometimes killing the executor-services, they've been moved to a separate process that can be given a larger Xmx.
While doing this, the ProcessMainClass was given utilities for the boilerplate surrounding receiving mq requests and responding to them, some effort was also put toward making the process boot process a bit more uniform. It's still a bit heterogeneous between different processes, but a bit less so for now.
Some refactoring is still needed, but an dummy actor is in place and a process that crawls URLs from the livecapture service's RSS endpoints; that makes it all the way to being indexable.
Also be a bit smarter about pre-allocating queues and sets based on depth rather than the number of provided URLs, which was always zero outside of tests.
