User-agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:68.0) Gecko/20100101 Thunderbird/68.6.0
Markus,
I believe that the server's general approach via download.php is
to consider the time stamp of the local file versus its timestamp
on the mirror. If they match, the mirror is a good one and will
be returned in the list. For the download folder, I expect the
timestamp is not so important because only individual files will
be downloaded from it. E.g., for this link, the timestamp of the
file is important, but not the timestamps of any of the parent
folders:
And here the timestamp of the 2020-03 folder itself is definitely
important.
That all being said, an install will only need to download 4
artifacts from the EPP repo, while needing to download 100s from
the train repo, so if there were no mirrors for the EPP repo, it
would not cause such a huge problem as it does when there are no
mirrors for the train repo. That repo already exists and has many
mirrors:
You are right. I used to be very careful with the
modification times of files and directories, especially the
download directories are important. I got used to using my
TIMESTAMP files that can be found here and there on the
download server, usually in commands like
touch -r . TIMESTAMP
touch -r TIMESTAMP .
to "store" the modification times of the current working
directory, and to use it afterwards to copy the very same
modification time back to it.
Of course it's not clear over which time period that
happened, but there's always quite a bit of activity the
day of and the days after... The mirror failures right
after the most 2019-12 release certainly brought that
download server to its knees enough to take note that
there was an ongoing problem. And that's in December,
close to Christmas time, when you wouldn't expect quite
so much either.
That's all to say that I very much appreciate Jonah
being conservative on this front.
Note too that I see that the mirrors are back now!
Now that I'm thinking about it though, we should be
careful what happens to the timestamp of the EPP folder
and the train folder when the content/artifacts change
just minutes before the release. If the folder's
timestamp changes, I believe all the mirrors will be
disabled again for a while...
Regards,
Ed
On 16.03.2020 16:42, Markus Knauer wrote:
Full agreement... 24h is typically enough.
Sometimes, if it was really urgent in the past, I
double-checked with Denis and the webmasters in
order to make them aware that there may be a
download peak. On the other hand, the peaks are not
that high (especially for a March release) than they
used to be.
And yes, I would definitely vote for the RC2a
version (without aeri).
I believe 24 hours is normally more
than enough, though the problem in https://bugs.eclipse.org/bugs/show_bug.cgi?id=561125
is still a problem so sooner is
probably better than later. I feel
that it's safe to assume that RC2a
versus RC2 will be good because the
difference is really very trivial and
I've confirmed that the Java, JEE and
RCP packages installed from this basis
via the installer start up fine...
Thanks Ed for confirming. I will
do the downloads this evening (Ottawa time)
Note that if you're going to transform
the EPP repo into a simple repository,
you should get the binary/features/plugins folders
copied up into the 2020-03 folder in
preparation as well.
I will place all the binary/features/plugins
files in place too, leaving just the
artifacts and content files to be updated
Wed morning.
I will ensure
that the composites are retuning 404 in time
for the release too.
Thanks,
Jonah
Thanks for all the heavy lifting!!
Regards,
Ed
On 16.03.2020 15:42, Jonah Graham
wrote:
Hi folks,
When do the files
need to be in place for release
day? I assume today is better than
tomorrow.
In that case we need
to have RC2a vs RC2 decision ASAP.
If I don't hear anything further I
will be making the RC2a files the
R files.