Are there any tuning profiles aimed at capsules

[mkearey]

Hi all, there appears to be no effort to tune capsule as we are for Satellites. Is this just considered unnecessary?

[jhutar]

Hello! No, we do not have that. Do you see a need for capsule tuning - in what scenario? Normally, capsules are not performance bottlenecks except for cases with too many Puppet classes/hosts.

[mkearey]

Greetings, we are looking at absolute performance of a satellite sync to a capsule. Mongo/pulp . We have no issues with puppet as we don't use it.

Cheers

[jhutar]

Hello. Only thing we have so far for Capsule is making sure it uses correct tuned profile:

https://github.com/RedHatSatellite/satellite6-performance-tuning/blob/master/docs/satellite-configuration-tuning.rst#tuned-profile

[psuriset]

@mccun934 can you please share, if we have come across any need to tune capsule? As Jan mentioned, Generic tuned profile on capsule has been good enough.

If we see any performance sensitive workloads on capsule, we would have to prioritize.

[pmoravec]

Just throwing some particular figures - rather to kick off some discussion (to correct my numbers :) ) than to conclude it:

• OS itself: 1 GB the most (in fact just 200MB-ish, but let round it to the upper bound)
• mongo: wiredTiger cache size: 4-6G, plus approx. 1GB for the rest, i.e. 5-7GB; might need more for some big repos and faster performance
• **qdrouterd, qpidd, squid, httpd, pulp_* (except for workers): up to 1GB**
• pulp workers: each can consume 1-2GB RAM (depending on size of processed repos and metadata); say 1.5GB per worker on average
• smart_proxy_dynflow_core + foreman-proxy: 1GB(?) might need more for scaled REX or frequent Hosts builds
• puppetserver: 2GB, not sure how much w need scaling here due to # of Hosts registered to the Capsule and due to complexity of puppet classes? I.e. can grow to 4GB on scaled deployments (no much experience from me, here).

In summary: 1+(5or7)+1+workers*1.5+(2or4) = (9 to 13) + workers*1.5 GB.

That is, for smaller deployments with 4 cores (hence 4 pulp workers) and following the lower values, I would recommend 9+4*1.5 = 15 GB RAM.

For larger deployments with 8 cores and upper bounds, 13 + 8*1.5 = 25 GB RAM.

Reviewing the final figures with reality I usually see in various deployments, I might over-estimate some figures (i.e. 12GB is often enough for small Caps, 20-24GB is often enough for a bigger one).

[Imaanpreet]

Hello, this time we come up with some capsule recommendations :

https://github.com/RedHatSatellite/satellite-performance-tuning/blob/devel/docs/satellite-configuration-tuning.rst#capsule-configuration-tuning

[Imaanpreet]

Added new updates to the capsule configuration tuning.

https://github.com/RedHatSatellite/satellite-performance-tuning/blob/f9862dca080748a6351824cadb8bfc866ff7c3e5/docs/capsule-configuration-tuning.rst

Thank you @jhutar @IshanSinghal .