Reduce disk writes

[ryandesign (Ryan Carsten Schmidt)]

The current mpbb/buildbot implementation causes large amounts of data to be written to disk, probably largely due to constantly activating and deactivating ports. This wears out disks that have a finite capacity to be rewritten. Can the amount of data written per build be reduced?

Four years ago I set up new buildbot workers with new SSDs, which through all the writing we've been doing have now worn out. To date they've endured between 440 and 540 terabytes written to each disk on SSDs only rated to accommodate 150TBW. 2 of the 3 SSDs have already fully failed and the third will likely fail soon as well. I'm replacing the SSDs but it would be nice if the replacements lasted longer.

For example, can we leave "all" ports activated and use trace mode? (We can't in fact leave all ports activated since some ports conflict with others, but we could for example leave non-conflicting ports activated.)

Or can activating a port be accomplished by a less expensive means than extracting an archive? MacPorts base used to install the destrooted files into hidden directories and activating the port consisted only of creating hard links to the already existing files. This method was later removed from MacPorts base but it had the advantage that it would write less data to disk.