meshagent-scrapy 0.39.9

Scrapy spider imports for Meshagent datasets

Meshagent Scrapy

Spider a website with Scrapy and import page content into a Meshagent room dataset.

from meshagent.scrapy import import_domain_with_scrapy

result = await import_domain_with_scrapy(
    room,
    url="https://example.com",
    table="pages",
    namespace=["crawls"],
    limit=100,
    concurrency=5,
)

To test it through meshagent room connect:

meshagent room connect --room=my-room --identity=scrapy -- \
  python meshagent-sdk/meshagent-scrapy/examples/crawl.py \
  https://www.meshagent.com --table=sample --namespace=crawls --limit=100 --concurrency=5

The sample command writes progress to stderr while it imports. TTY output uses a single updating line; redirected output uses plain log lines. Pass --silent to suppress progress output.

Pass --concurrency or concurrency= to tune Scrapy's maximum concurrent requests.

Pass --batch-size or batch_size= to cap how many page records are merged into the content table at once. The crawler also flushes content batches by estimated payload size with --max-batch-bytes or max_batch_bytes=, which defaults to 16 MiB, and by elapsed time with --max-batch-delay or max_batch_delay=, which defaults to 60 seconds. The row-count cap defaults to 100. Raw HTML rows can be large, so prefer lowering the byte limit before lowering the row count if the room server reports Lance/DataFusion merge memory exhaustion while importing full HTML pages.

The crawler sends a browser-like User-Agent by default. Pass --user-agent or user_agent= to override it for a specific crawl.

The default extractor writes page content as markdown in the text column. Use --format=html to keep HTML, --format=text to strip markup to plain text, or pass content_format= from library code.

By default, the crawler runs Trafilatura cleanup before converting markdown/text content, which strips common navigation, footer, sidebar, and ad boilerplate. For --format=html, the default is to strip scripts and inline image data URLs while preserving the rest of the HTML. Use --strip= with comma-separated values like scripts, css, whitespace, image-data-urls, or clean to choose the HTML stripping steps, or --strip=none to process the raw response body.

The CLI persists crawl frontier state by default in <table>__frontier, so a limited run can be resumed by running the same command again:

meshagent room connect --room=my-room --identity=scrapy -- \
  python meshagent-sdk/meshagent-scrapy/examples/crawl.py \
  https://www.meshagent.com --table=sample --namespace=crawls --limit=100

Pass --frontier-table to choose a different state table, or --no-resume to run without frontier persistence. Library callers can opt in with resume=True. Frontier updates are buffered before they are written; tune that with --frontier-batch-size or the library frontier_batch_size= argument. Failed URLs are not retried on resume unless you pass --retry-failed or retry_failed=True.

The crawler creates indexes by default: a BTREE index on the page table primary key, plus BTREE url and BITMAP status indexes on the frontier table. Pass --index=text or index_columns=("text",) to also create an INVERTED index on text. Pass --no-indexes or create_indexes=False to skip all automatic index creation. It also runs dataset optimization periodically while importing and shows optimizing/optimized in progress output. Tune that with --optimize-every or optimize_every=, and use 0 on the CLI or None in library code to disable automatic optimization.

By default, the crawler imports textual responses only, based on Content-Type values containing text/, html, xml, or json. Pass --response-filter or response_filter= to replace that default with a JMESPath expression over url, status, headers, content_type, and content_type_lower. Header names are lower-cased, so an HTML-only crawl can use:

--response-filter "contains(headers.\"content-type\", 'text/html')"

By default, records are merged on url with the columns url, date, content_type, text, and images. text is markdown unless another content format is selected. images is a struct array with src and alt only, and inline image data URLs are excluded.

Pass an async extract= callback to derive custom columns from the Scrapy response and content bytes. Return None from the callback to skip the record. Pass an async progress= callback to observe import progress from library code.