kke
commented
6 years ago The labelSelector can do in queries, so it would be possible to just one big prune that does one query if all the phases and addons would return a list of stacks + checksums they applied?
Running the pruning in a background thread while the next step is already running does not seem to shave off very much from the total execution time.
==> Reading instructions ...
==> Sharpening tools ...
==> Starting to craft cluster ...
==> Validate hosts @ 192.168.100.100 192.168.100.101 192.168.100.102
==> Configure hosts @ 192.168.100.100 192.168.100.101 192.168.100.102
==> Configure kubelet @ 192.168.100.101 192.168.100.102
==> Configure master @ 192.168.100.100
==> Configure kube client @ 192.168.100.100
==> Configure DNS @ 192.168.100.100
==> Configure network @ 192.168.100.100
pruning started for 4f2c5d5868b2c7f39fe8e4b3d5ccd50c
pruning ended for 4f2c5d5868b2c7f39fe8e4b3d5ccd50c, took 0.4555
==> Configure metrics @ 192.168.100.100
pruning started for 188d95bbe1f098a2dd3cd99a754a2fdc
pruning ended for 188d95bbe1f098a2dd3cd99a754a2fdc, took 0.3148
pruning started for d4d36e14048141fdd32ab0da0d60d118
pruning ended for d4d36e14048141fdd32ab0da0d60d118, took 0.2916
==> Store cluster YAML @ 192.168.100.100
==> Configure bootstrap tokens @ 192.168.100.100
==> Join nodes @ 192.168.100.101 192.168.100.102
==> Label nodes @ 192.168.100.100 192.168.100.101 192.168.100.102
==> Configuring addons ...
==> Enabling addon ingress-nginx
pruning started for 533650c1533b14bf5545295133db8bce
pruning ended for 533650c1533b14bf5545295133db8bce, took 0.2929
==> Enabling addon host-upgrades
pruning started for 5bb109ab84690d0f158c2c56bfed98e1
pruning ended for 5bb109ab84690d0f158c2c56bfed98e1, took 0.2904
==> Disabling addon cert-manager
pruning started for -
pruning ended for -, took 0.2983
==> Disabling addon kured
pruning started for -
pruning ended for -, took 0.3333
==> Disabling addon kubernetes-dashboard
pruning started for -
pruning ended for -, took 0.3068
==> Cluster has been crafted! (took 16 seconds)Same with background pruning:
==> Reading instructions ...
==> Sharpening tools ...
==> Starting to craft cluster ...
==> Validate hosts @ 192.168.100.100 192.168.100.101 192.168.100.102
==> Configure hosts @ 192.168.100.100 192.168.100.101 192.168.100.102
==> Configure kubelet @ 192.168.100.101 192.168.100.102
==> Configure master @ 192.168.100.100
==> Configure kube client @ 192.168.100.100
==> Configure DNS @ 192.168.100.100
==> Configure network @ 192.168.100.100
==> Configure metrics @ 192.168.100.100
pruning started for 25fba6be91a38b1eb271a586f39ef684
pruning started for 759db0d5eca26a10a2e4e6a2af989c22
pruning ended for 25fba6be91a38b1eb271a586f39ef684, took 4.8879
==> Store cluster YAML @ 192.168.100.100
pruning started for ebbcbaccf55cacb29d5590f4851ed492
==> Configure bootstrap tokens @ 192.168.100.100
pruning ended for 759db0d5eca26a10a2e4e6a2af989c22, took 2.9185
==> Join nodes @ 192.168.100.101 192.168.100.102
==> Label nodes @ 192.168.100.100 192.168.100.101 192.168.100.102
==> Configuring addons ...
==> Enabling addon ingress-nginx
pruning ended for ebbcbaccf55cacb29d5590f4851ed492, took 0.6649
==> Enabling addon host-upgrades
pruning started for c30184507c9a3865305caef88e191d72
==> Disabling addon kured
pruning started for -
pruning started for 693b0ea33c745fb0e25c104fd6ff0bf9
pruning ended for c30184507c9a3865305caef88e191d72, took 0.6068
pruning ended for 693b0ea33c745fb0e25c104fd6ff0bf9, took 0.5827
pruning ended for -, took 0.5962
==> Disabling addon kubernetes-dashboard
pruning started for -
pruning ended for -, took 0.3172
==> Disabling addon cert-manager
pruning started for -
pruning ended for -, took 0.3221
==> Cluster has been crafted! (took 15 seconds)
SpComb
Timing it manually, each stack takes roughly ~10s to apply.... I think most of that is in the
prune_stack, and all theget_entitiesrequests.Each
pharos-cluster updoes 8 stack applies (weave + heapster/metrics-server + 5 addons)... if a no-oppharos-cluster upwith #111 for parallel SSH nodes takes about 90s, then the majority of that time is spent pruning stacks.Request log with
RESTCLIENT_LOG=stdout: