Premier League Predictor: FastAPI & Redis — Chapter 9, Exercise 3
=================================================================

TASK
Start a default Redis container and one with the append-only file and
fsync always, write two keys to each, kill both with docker kill, restart
them and count the keys. Then repeat the default one with docker stop.
Report all three results and say which restart behaviour to rely on for
hand-entered fixtures.

SOLUTION

  docker run -d --name plredis -p 6399:6379 redis:latest
  docker run -d --name plaof   -p 6398:6379 redis:latest \
         redis-server --appendonly yes --appendfsync always

Config of the default container (CONFIG GET):

  save        = 3600 1 300 100 60 10000
  appendonly  = no
  appendfsync = everysec

Both containers: SET season:current_id 7, and HSET fixture:1 with
home_score 2, away_score 1 (DBSIZE = 2).

  docker kill plredis plaof ; docker start plredis plaof

  default:            DBSIZE 0   season:current_id -> None
  appendonly always:  DBSIZE 2   season:current_id -> '7'

Default container again, written to and then stopped gracefully:

  docker stop plredis ; docker start plredis
  default:            DBSIZE 2   season:current_id -> '7'

Results:
  - default + hard kill      -> everything lost (no snapshot had been taken yet)
  - default + graceful stop  -> kept (shutdown writes a snapshot)
  - AOF + fsync always + hard kill -> kept

Which to rely on: the fixtures and predictions in this app are typed in
by hand and cannot be recomputed from anything else, so a hard kill must
not lose them. That points to the append-only file (accepting the cost of
an fsync on each write, or choosing a setting between the two and
understanding the window it leaves), not to the defaults. The table and
leaderboard are derived data, so losing them is recoverable; losing
fixtures is not.

WHY THIS WORKS AS AN ANSWER
----------------------------
It reproduces all three behaviours with real restarts, states exactly what
was and was not tested, and ties the choice to which data is source data
and which can be rebuilt.