| The gc query was already similar, though not the same as the newer fact-contents query, and testing the existing gc query on a fast machine against a version of the fact-contents query that had been adjusted to only track the relevant data showed that the existing gc query was already roughly twice as fast. Though for context, that was on a very fast nvme drive, and while it was running against 100k nodes, they were benchmark generated, and not rewritten, so that table was presumably very compact, etc. Since we've seen that gc query taking 20+ minutes at client sites with fewer nodes, we'd recommend working with CS to gather some relevant information from a collection of notable sites as a next step – both data for which we'd provide the collection tools, and the coincident support script (i.e. sar) data. |
