Change Details

According to https://w.wiki/663g (note: sample rate is 1:128), [https://github.com/openzim/mwoffliner|MWOffliner]] hits us with about 7000 requests per minute for the action=parse API, apparently to retrieve a rendered version of a page for offline use, along with associated meta data. That's roughly 1/4 of the traffic on that API (total is 31k) This is not an immediate problem, but seems inefficient. It also causes a lot of writes to the parser cache which may otherwise not be needed (*). This seems like a good use case for MWE, or at least the new page/{title}/html REST API. (*) needs more investigation; the rate of ParserCache writes caused by api_parse is 18k/minute, so this is the cause of 1/3. It's not clear whether all 7000 requests from MWOffliner result in a cache miss, though.

According to https://w.wiki/663g (note: sample rate is 1:128), [https://github.com/openzim/mwoffliner|MWOffliner]] hits us with about 7000 requests per minute for the action=parse API, apparently to retrieve a rendered version of a page for offline use, along with associated meta data. That's roughly 1/4 of the traffic on that API (total is 31k) This is not an immediate problem, but seems inefficient. It also causes a lot of writes to the parser cache which may otherwise not be needed (*). This seems like a good use case for MWE, or at least the new page/{title}/html REST API. (*) needs more investigation; the rate of ParserCache writes caused by api_parse is 18k/minute, which would be about 1/3 of the cache writes. It's not clear whether all 7000 requests from MWOffliner result in a cache miss+write, though. It seems likely that many of them are cache hits.

According to https://w.wiki/663g (note: sample rate is 1:128), [https://github.com/openzim/mwoffliner|MWOffliner]] hits us with about 7000 requests per minute for the action=parse API, apparently to retrieve a rendered version of a page for offline use, along with associated meta data. That's roughly 1/4 of the traffic on that API (total is 31k) This is not an immediate problem, but seems inefficient. It also causes a lot of writes to the parser cache which may otherwise not be needed (*). This seems like a good use case for MWE, or at least the new page/{title}/html REST API. (*) needs more investigation; the rate of ParserCache writes caused by api_parse is 18k/minute, so this iswhich would be about 1/3 of the cause of 1/3che writes. It's not clear whether all 7000 requests from MWOffliner result in a cache miss+write, though. It seems likely that many of them are cache hits.