Hello Mark, Mark H Weaver <m...@netris.org> skribis:
> It's a good question. I have several hypotheses: These are all valid but there’s a couple more to consider. ;-) Specifically we’ve had ENOSPC issues on some build nodes lately, and as I wrote elsewhere, ‘guix offload’ would report them as “permanent failures”. Thus guix-daemon on berlin would cache those failures and never retry afterwards. This is fixed by commit b96e05aefd7a4f734cfec3b27c2d38320d43b687. Commit 63b0c3eaccdf1816b419632cd7fe721934d2eb27 also arranges so we don’t choose machines low on disk space. Another issue I’ve noticed is “database is locked” offloading crashes, fixed by bdf860c2e99077d431da0cc1db4fc14db2a35d31. We probably don’t get these on hydra.gnu.org because we’re running a version that predates the replacement of the ‘guix-register’ program by (guix store database). There’s a few more issues about offloading in the bug tracker. I suspect these explain the low availability of substitutes to a large extent. Ludo’.
