Summary
I deployed the fluxer container via docker and after 2 weeks i got high Net I/O and Block I/O readings. On the first day I connected to the voice channel a maximum of 1h with a friend. Then forgot about it 2 weeks and I was checking on a different matter the container stats withdocker statsSteps to reproduce
- set up a self-hosted container with docker
- set up a server
- watch the Net I/O and Block I/O increase with the command
docker statswithout causing obvious traffic
Logs or screenshots
docker statsdocker stats

6 comments
Comment by @ToxykAuBleu
Comment by @UnknownLITE
631195939-998bb92d-6019-4594-8331-a3745575d599.png
1110×493 | 104 kB
Comment by @SIGTERM-015
pg_stat_statementsand caught the exact statement. Numbers below are from 5 minutes on a completely idle instance (no users connected, no activity), counters reset immediately before. Top statement by rows returned:row_keypredicate, so this loads an entire logical table in one go. 278,266 is exactly the row count ofdisposable_email_domainsin my instance:jobs_by_idandjobs_by_day_bucket(126,649 each). That accounts for the 20.3 GB of block writes over 6 days. Two separate problems, then:- A full-table load of a large static dataset on a short timer. The blocklist is reference data that changes rarely — it looks like a cache refresh that re-reads all 278k rows instead of loading once, watching for changes, or paging.
- An idle job pipeline writing ~136k rows/day. Something enqueues work continuously with no users present.
Both are amplified by the single-heapfluxer_kvdesign: any sequential scan pays for unrelated datasets sharing the same physical table.Comment by @kerichdev
Comment by @MrRubberDucky
podman stats(containers were up for 10 hours)Comment by @Z3phyr1s