psiphon: test bootstrap with pre-existing cache

[bassosimone]

We always bootstrap from scratch. But this prevents us from understanding whether a pre-existing Psiphon cache could actually allow us to perform measurements. We can do better by keeping a cache on disk and doing both a cache-less and a cache-full bootstrap. This issue is labeled "data quality" because it will probably increase the quality of our data to add this functionality. This issue also requires research to understand the implications of reusing an existing cache.

[bassosimone]

I asked to Psiphon developers for clarifications around the implications of killing the cache before running each test.

[bassosimone]

Okay. I wrote a minimal tool to check what happens with a pre-existing cache. There's no need to reveal the source, since it's just a very stripped down version of the code that we use in ooni/probe-cli. Here's the output of a few runs:

$ ./psitest 
{"Address":"127.0.0.1:36095\n","BootstrapTime":5.174298996,"DirExists":false}

$ ./psitest -keep
{"Address":"127.0.0.1:32921\n","BootstrapTime":5.43916411,"DirExists":false}

$ ./psitest -keep
{"Address":"127.0.0.1:36069\n","BootstrapTime":0.587030723,"DirExists":true}

$ ./psitest -keep
{"Address":"127.0.0.1:44917\n","BootstrapTime":0.595344917,"DirExists":true}

$ ./psitest -keep
{"Address":"127.0.0.1:39079\n","BootstrapTime":2.251123913,"DirExists":true}

$ ./psitest -keep
{"Address":"127.0.0.1:44307\n","BootstrapTime":0.398089054,"DirExists":true}

So, the time to bootstrap is really faster when using a cached directory. We need to better investigate how we're managing the Psiphon cache in OONI because I suspect we're doing something wrong in probe-cli.

[bassosimone]

I also noticed that when OONI is running Psiphon this does not happen. This means that, most likely, there is something wrong in our code such that we are not actually using the Psiphon cache. This explains the slow bootstrap 🤦.

[bassosimone]

We can close this issue now. The investigation is clearly complete.