On Thu, Jun 15, 2023 at 9:01 AM Boris Behrens <[email protected]> wrote:
> Hi Ben, > I've read the link you gave me and I will try to answer the best I can. > > > What problem are you solving for? > I want to keep historical data for longer periods with less disk space > used. > I currently keep around 4 month of data for around 1.3T of disk space. > and I would love to have 1-2 years of data without having to buy 20TB > disks :) > 2 years of data in your case is only 8T. That's not a large amount of disk. > > > Why do you think you need this? > For debugging and monitoring I only need the last couple of days with > a raw resolution. And seeing trends can be downsampled (is this the > correct term?) because it is just historical trend data. > I think it's rather a use case than an actual problem I am working on. > I would like to see in the past how things changed and overlay it with > timestamps from changes. Things begin to looks way different when you > check the past. > Yes, downsampling is the correct term. But downsampling is not required to do this. Grafana supports special `$__interval` and `$__rate_interval` variables that can be used to automatically scale graphs based on desired time ranges. ` > > You say "a lot of endpoints", but what is a lot? 1,000? 10,000? 100,000? > 1,000,000? How many series? > A couple hundred endpoints with around 1.5m of series. But I don't > think that this information is helpful in the current context. > It is, because it gives us a sense of scale for your problem. You're running a small deployment, so no fancy changes are really necessary. For comparison, my $dayjob has about 900TiB of TSDB storage (Thanos object storage). Managing this is a different scale to a single node deployment. I think you're falling into the premature optimization trap. You could use a full Thanos deployment, as it supports exactly the kind of variable retention downsampling you want. But it's going to take a lot more work than just buying slightly bigger disks to store Prometheus on. > > Thank you for your time and effort to help me with my stupid little > problem. > Cheers > Boris > > Am Do., 15. Juni 2023 um 06:24 Uhr schrieb Ben Kochie <[email protected]>: > > > > My first question is "why?" (https://xyproblem.info/) > > > > What problem are you solving for? Why do you think you need this? > > > > You say "a lot of endpoints", but what is a lot? 1,000? 10,000? 100,000? > 1,000,000? How many series? > > > > On Thu, Jun 15, 2023 at 6:21 AM Boris <[email protected]> wrote: > >> > >> Dear prometheus community, > >> > >> my prometheus scrapes a lot of endpoints and I only need to 10s for a > month. After that I would like to rollup the data to 5m averages and after > 6 months I would like to have only hourly averages. > >> > >> So I can have quite a long time of data and can see changes over a long > time with the same dashboard. > >> > >> I am sorry to ask this question but it seems like I don't have the > correct search term. Downsampling seems to keep the raw data and all > searches lead me to timescaledb. > >> > >> Cheers > >> Boris > >> > >> -- > >> You received this message because you are subscribed to the Google > Groups "Prometheus Users" group. > >> To unsubscribe from this group and stop receiving emails from it, send > an email to [email protected]. > >> To view this discussion on the web visit > https://groups.google.com/d/msgid/prometheus-users/8D01579C-5696-40BB-9841-A1E3E0094DD1%40kervyn.de > . > > > > -- > Die Selbsthilfegruppe "UTF-8-Probleme" trifft sich diesmal abweichend > im groüen Saal. > -- You received this message because you are subscribed to the Google Groups "Prometheus Users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/prometheus-users/CABbyFmojP%2BsB4%2BdwGV%3DnJkSp-zVa%3Dn%2BXz5Qd1i3eOgAy%3DU_tsw%40mail.gmail.com.
