The Cambridge LAVA Lab has a transparent squid proxy which saves having to configure each dispatcher and device to use it. Outgoing HTTP traffic from the lab has no choice as it is intercepted at the internet gateway.
We did this because even after configuring the dispatcher, and devices, it's almost impossible to make all test shell tasks use the proxy. LAVA sets a shell environment inside a job but many of the clients in the various different types of job simply ignore it. Chasing every test writer was not feasible as the lab usage is so large, but might be ok in a smaller lab with tighter control of the jobs.
We don't proxy HTTPS requests because that's becomes very complicated with faking certificates etc
Matt Hart matthew.hart@linaro.org writes:
In kernel CI, the downloads required to get a job started (kernel, dtb, ramdisk etc.) often take longer than running the job itself, so I primarily want the dispatcher downloads to go through a proxy.
I' don't care if the targets and tests themselves go throught the proxy, so fixing LAVA to use the proxy from the environment is still useful, IMO.
Kevin
Oh agreed, was just saying how/why the lab does it transparently
On 14 April 2016 at 18:01, Kevin Hilman khilman@baylibre.com wrote:
lava-users@lists.lavasoftware.org
-
Kevin Hilman -
Marc Titinger -
Matt Hart