Call repo_check_timeout() before colleting the POLLOUT fds. Since
repo_abort() called by repo_check_timeout() will add messages to
be sent out.
This brings back rev 1.263 which was accidentially reverted by rev 1.293
OK tb@
geo/py-owslib: add missing RDEP on py-lxml
OGC protocols being mostly XML-based, it somewhat makes sense to depend
on an XML parser...
found by Marco Van Hulten via the MetaSearch QGIS plugin
When the pagedaemon is triggered to create free memory, there may be
sleeping pmemrange allocations with multi-page alignment requirements
which can't be satisfied by the simplistic freeing of (solo) pages
which the pagedaemon performs. As we near starvation, fragmentation
is the main problem. Our free list could be large enough that the
pagedaemon sees no reason to do more work, but also too fragmented to
satisfy a pending allocation request with complex requirements
(imagine asking for 512K of physically linear memory which is DMA
reachable). When the requirement isn't satisfied, the pagedaemon is
told to try again, but again doesn't mean harder because it has no
mechanism to try harder. It's tracking variables do not show the
fragmentation problem. It spins a lot. Often this becomes a
deadlock.
Time to change strategy: Overshoot creation of (both) inactive and
free pages each time through the loop. After inspecting existing
variables, we generate minumum 128 inactive pages (which may be
dynamically drawn down asyncronously by accesses), and then try to
convert minumum 128 inactives into free pages (different pages
get freed different ways, including via swapcluster which has been
[7 lines not shown]
To support swapencrypt, the swapcluster code has a memory allocation codepath.
Since this is runs inside the pagedaemon that is unworkable. We'd like to
encrypt the pages inplace for IO, but there are architectures not ready for
a high-mem page to be written to a dma-restricted device (work in progress).
So for now we need to bounce through dma-reachable memory buffer. A previous
attempt had 1 extra bounce buffer, but then slept on allocation inside the
pagedaemon context which is also unworkable. This version contains 32
pre-allocated swapclusters (64K each), and through a counter signals to the
pagedaemon when it should stop trying to create memory. 32 swap clusters
is comfortably more than the minimum we expect the pagedaemon frantically
generate. This crummy solution is good enough until we the dma reach problem
is solved (soon)
ok kettenis kirill (who looked into other solutions) beck
