Today I noticed that osmo-msc-master was being built several times in a row and
I found there are multiple redundant ricocheting downstream builds of the same
original upstream build timer.
All of:
https://jenkins.osmocom.org/jenkins/job/master-osmo-msc/18894/https://jenkins.osmocom.org/jenkins/job/master-osmo-msc/18895/https://jenkins.osmocom.org/jenkins/job/master-osmo-msc/18896/
show:
Started by upstream project master-libosmocore build number 1624
And then there are:
https://jenkins.osmocom.org/jenkins/job/master-osmo-msc/18897/https://jenkins.osmocom.org/jenkins/job/master-osmo-msc/18898/
showing:
Started by upstream project master-libosmo-abis build number 2611
So there are two timers launching master-libosmocore and launching
master-libosmo-abis, and each of these cause multiple downstream build triggers
of osmo-msc. Then there is also the osmo-msc timer, independently launching
more nightly master-osmo-msc builds.
It is not necessary to depend these builds on each other, because none of them
depends on artifacts built in another job, each of them in turn builds the
entire chain of libosmocore up to that build's target project.
Upon a libosmocore *code change*, it makes sense to trigger downstream builds,
but not when caused by a timer or for other jenkins trigger or user reasons.
I vaguely remember discussing this before, but now I realize we could solve
this and have less redundant builds:
Do not configure libosmocore to trigger downstream builds of
master-libosmo-abis (and so on), but instead trigger master-libosmo-abis upon
each libosmocore patch merge. That can be done by adding a "Gerrit trigger" on
the master-libosmo-abis build job config, making a patch-merge to the gerrit
libosmocore.git cause a master-libosmo-abis build. At that point, jenkins
timers no longer cause downstream builds, but merges to master on upstream git
repositories do still cause builds.
(maybe there is also a normal git SCM way that doesn't require
gerrit.osmocom.org?)
For projects like osmo-msc, we would add multiple projects as upstream git
repositories, like libosmocore, libosmo-abis, libosmo-netif, ... , osmo-iuh,
osmo-mgw, ...
Basically all of the projects with more than one dependency get retriggered
redundantly every night, some doing their entire build matrix over several
times for no good reason.
I'm not sure that it's really worth the effort, but I thought I'd mention it
since seeing so many redundant builds on various osmo source trees at night
seems avoidable stupidity. Should someone spend cycles to implement this?
~N
Hi all,
as some of you may know, I am currently working on baseband frequency
hopping implementation for osmo-bts-trx (see OS#4546). I've already
submitted more or less final implementation to Gerrit [1], and in
general it works fine [2].
The idea is quite simple: you have several RF carriers (transceivers)
with fixed frequencies, so on hopping timeslots osmo-bts-trx distributes
bursts between them according to the configured hopping sequence.
However, when it comes to hopping timeslots on C0, the burst
distribution algorithm gets a bit more complicated.
The problem is that on C0 the BTS has to maintain constant power level,
even if there is nothing to transmit, so the phones can detect it
reliably during the power scan. This is achieved by sending dummy
bursts. When frequency hopping is not in use, the scheduler in
osmo-bts-trx sends a dummy burst immediately. When it is in use,
sending dummy bursts needs to be postponed until all transceivers are
processed, because some other transceiver might need to send a normal
burst on the frequency of C0, and there would be no need to send a dummy
burst in that case. So I came up with [3] (any comments regarding
potential optimizations and further improvements are welcome).
While working on this, I stumbled upon the burst filling feature in
osmo-trx [4], that is disabled by default. If enabled, osmo-trx would
be sending dummy bursts itself, so there is no need to send them from
osmo-bts-trx, and thus no need to complicate the burst distribution logic.
I am now wondering why don't we enable this feature by default?
From the VTY reference:
> Send a dummy burst when there is nothing to send on C0 (TRX0) and
> empty burst on other channels. Use for OpenBTS compatibility only,
> don’t use with OsmoBTS as it breaks encryption.
it's unclear how sending dummy bursts would break encryption...
[1] https://gerrit.osmocom.org/c/osmo-bts/+/19030
[2] https://www.youtube.com/watch?v=tGMVNYo9s7E
[3] https://gerrit.osmocom.org/c/osmo-bts/+/19029
[4]
http://ftp.osmocom.org/docs/latest/osmotrx-vty-reference.pdf#subsection.1.9…
Best regards,
Vadim.
--
- Vadim Yanitskiy <vyanitskiy at sysmocom.de> http://www.sysmocom.de/
=======================================================================
* sysmocom - systems for mobile communications GmbH
* Alt-Moabit 93
* 10559 Berlin, Germany
* Sitz / Registered office: Berlin, HRB 134158 B
* Geschaeftsfuehrer / Managing Director: Harald Welte
