There were a couple places in Google's internal infrastructure where they had picked up traditional "big iron" architectures (mostly by acquisitions), and it was funny to observe the sheer plane between their regular best practices and those systems (when they hadn't yet been replaced).
One high-ticket item relied on a huge Oracle database. Every other aspect of that product was patch-on-the-fly, silent-release new versions, forward-and-backwards compatibility of software... But the whole product had regular scheduled outages on the weekends because the Oracle part could not be patched on the fly, had a rigid schema baked into its relational architecture, and had to be brought offline for updates and migrations.
It was like everyone was zipping around in racecars and then they all had to pause for Grandpa to cross the street.
One high-ticket item relied on a huge Oracle database. Every other aspect of that product was patch-on-the-fly, silent-release new versions, forward-and-backwards compatibility of software... But the whole product had regular scheduled outages on the weekends because the Oracle part could not be patched on the fly, had a rigid schema baked into its relational architecture, and had to be brought offline for updates and migrations.
It was like everyone was zipping around in racecars and then they all had to pause for Grandpa to cross the street.