Age | Commit message (Collapse) | Author |
|
svn path=/branches/tk705/; revision=6344
|
|
is now handled in the outermost script, before rpki.rpkic is even
loaded; rpki.rpkic just assumes that euid and ruid have been set up
correctly and swaps them temporarily when it needs to do file I/O.
svn path=/branches/tk705/; revision=6340
|
|
GUI no longer uses (an additional layer of) temporary files between
itself and zookeeper. Zookeeper file read methods now take file-like
objects. rpkic now opens input files as the real UID, then reverts
back to the effective UID before handing the resulting file off to the
zookeeper.
This caught several more instances of rpkic file I/O that were not
doing the real/effective UID swap properly while loading CSV and VCard
files.
As far as I can tell from testing, this didn't break anything that
worked before. Whether it fixed all the file I/O problems remains to
be seen.
svn path=/branches/tk705/; revision=6339
|
|
rpki.irdb.zookeeper.etree_read(), the functions that call it, and the
functions that call them, could all use a bit of refactoring. At this
point pretty much ever caller is jumping through some hoop or another
due to the way the code has evolved, and most of it could be simpler.
svn path=/branches/tk705/; revision=6338
|
|
more interesting way now.
svn path=/branches/tk705/; revision=6335
|
|
cron job stuff while we're at this.
svn path=/branches/tk705/; revision=6334
|
|
Django settings with those of all the other ORM-using programs.
svn path=/branches/tk705/; revision=6333
|
|
svn path=/branches/tk705/; revision=6330
|
|
remove all the other copies to prevent confusion.
svn path=/branches/tk705/; revision=6329
|
|
stop recent Debian and Ubuntu from whining at the user on
installation.
svn path=/branches/tk705/; revision=6327
|
|
other engines.
svn path=/branches/tk705/; revision=6326
|
|
"unstable" where there is no numeric version available (yet, or ever,
as the case may be).
Debian Stretch picked up the mandatory ".conf" suffix for Apache site
files; don't (yet) know whether that was already present in Jessie,
won't much care until we have working backports of our dependencies
for Jessie.
svn path=/branches/tk705/; revision=6325
|
|
svn path=/branches/tk705/; revision=6324
|
|
when RPKI_USER not found.
svn path=/branches/tk705/; revision=6323
|
|
svn path=/branches/tk705/; revision=6322
|
|
rcynic-cron.
NB: rpkigui-rcynic is currently broken and is therefore disabled in rcynic-cron.
This needs to be fixed, but it was broken on this branch in any case, no real
point in holding up the rest of this waiting for it.
svn path=/branches/tk705/; revision=6321
|
|
of having to include them as build dependencies just so that
./configure can figure out their locations.
svn path=/branches/tk705/; revision=6320
|
|
svn path=/branches/tk705/; revision=6319
|
|
svn path=/branches/tk705/; revision=6318
|
|
svn path=/branches/tk705/; revision=6317
|
|
svn path=/branches/tk705/; revision=6316
|
|
rpki.autoconf gets the right values. Better fix would be to set these
on the ./configure command line in debian/rules, try that later.
svn path=/branches/tk705/; revision=6314
|
|
control over precisely what this rsync client is allowed to do.
svn path=/branches/tk705/; revision=6313
|
|
svn path=/branches/tk705/; revision=6312
|
|
svn path=/branches/tk705/; revision=6310
|
|
current RRDP specification, and get the URI right while we're at it.
svn path=/branches/tk705/; revision=6309
|
|
svn path=/branches/tk705/; revision=6308
|
|
svn path=/branches/tk705/; revision=6307
|
|
svn path=/branches/tk705/; revision=6306
|
|
svn path=/branches/tk705/; revision=6305
|
|
back the way they were before we started with XMLPullParser.
svn path=/branches/tk705/; revision=6304
|
|
cleanup, but runs more than an order of magnitude faster in common
case where the vast majority of objects in the snapshot are new, and
(in theory) should not be significantly slower than what we were doing
already in more complicated mixtures.
svn path=/branches/tk705/; revision=6303
|
|
svn path=/branches/tk705/; revision=6302
|
|
svn path=/branches/tk705/; revision=6301
|
|
svn path=/branches/tk705/; revision=6300
|
|
quickly when processing large snapshots which take (much) longer to
load into SQL than to pull down from the net. Given this change,
lxml.etree.iterparse() is a more suitable API than
lxml.etree.XMLPullParser, so switch.
This version can download and process a 120,000 object snapshot,
albeit hideously slowly.
svn path=/branches/tk705/; revision=6299
|
|
svn path=/branches/tk705/; revision=6298
|
|
svn path=/branches/tk705/; revision=6297
|
|
svn path=/branches/tk705/; revision=6296
|
|
using a single bulk IRDB query for all the child data that
UpdateChildrenTask needs is a better solution.
svn path=/branches/tk705/; revision=6295
|
|
svn path=/branches/tk705/; revision=6294
|
|
svn path=/branches/tk705/; revision=6293
|
|
svn path=/branches/tk705/; revision=6292
|
|
already been postponed at least once, we stay with the current task
rather than doing a forced round-robin. This may require further
tweaks, eg, to flush the publication queue at regular intervals even
if we're not postponing, but the intent is to allow tasks with complex
setup to run to completion while remaining responsive, to avoid
wasting all of our time on repeated setup overhead.
svn path=/branches/tk705/; revision=6291
|
|
svn path=/branches/tk705/; revision=6290
|
|
svn path=/branches/tk705/; revision=6289
|
|
both this and certificate serial number allocation under a lock, but
doing so would require converting a lot of plain methods into
coroutines, so postpone that decision until we're done with revisions
to the task scheduler.
svn path=/branches/tk705/; revision=6288
|
|
svn path=/branches/tk705/; revision=6287
|
|
certificate, to make certain that it is not later than manifest
thisUpdate value.
svn path=/branches/tk705/; revision=6286
|
|
svn path=/branches/tk705/; revision=6285
|