Getting started
You need the CLI and the CRD at every site (Install), and a Workload file that names the sites, their kube contexts and the Git repository (the Workload file).
A planned switchover
With a Workload file for sites a and b, a planned switchover from a to b:
sh
# Read every site and Git, print every step, change nothing.
tack plan -f workload.yaml --to b
# Print the plan again, ask for the workload's name, then run it.
tack run -f workload.yaml --to b
# List the workload's runs at every site, or show one run step by step.
tack status -f workload.yaml
tack status pg-main-20261027-101500 -f workload.yaml
# Continue a run that stopped (exit code 3) or waits (exit code 4).
tack resume pg-main-20261027-101500 -f workload.yaml
# Delete the ledgers of old runs that succeeded, keeping the 20 newest of them.
tack prune -f workload.yaml --keep 20After a site loss, add --mode site-loss to plan and run, after the checks of Fail over after a site loss. Running Tack covers stopped runs and break-glass.
What a plan shows
An excerpt of tack plan:
text
Plan for workload pg-main: Planned, site a to site b
Ledger: SiteFailover tack-system/pg-main-20261027-101500 at site b
Sites:
a Primary declares a healthy Cluster db/pg-a: phase "Cluster in healthy state", declares pg-a primary, pod pg-a-1; Pooler pg-a-pooler-rw has 2 instances
b Replica declares a healthy Cluster db/pg-b: phase "Cluster in healthy state", declares pg-a primary, pod pg-b-1; Pooler pg-b-pooler-rw has 0 instances
...
Failover
...
9. demote-a site a
demote Cluster pg-a: declare pg-b primary and drop any old token
patch Cluster db/pg-a (merge): {"spec":{"replica":{"primary":"pg-b","promotionToken":null}}}
10. wait-demoted-a site a
wait until Cluster pg-a has demoted and published its demotion token
wait up to 15m0s on Cluster db/pg-a until:
{.status.demotionToken} is not empty
{.status.switchReplicaClusterStatus.inProgress} NotEquals "true"
11. capture-token site a
record the demotion token of Cluster pg-a in the ledger
record {.status.demotionToken} of Cluster db/pg-a in the ledger as $(demotionToken)
12. promote-b site b
promote Cluster pg-b with the demotion token, in one patch
patch Cluster db/pg-b (merge): {"spec":{"replica":{"primary":"pg-b","promotionToken":"$(demotionToken)"}}}
...
PromoteToGitOps
18. git-commit Git
commit the live state to ssh://***@git.example.com/gitops.git branch main
clusters/site-a/db/cluster.yaml, document 0: spec.replica.primary = "pg-b" (site a, Cluster db/pg-a spec.replica.primary)
clusters/site-a/db/cluster.yaml, document 0: spec.replica.promotionToken removed (site a, Cluster db/pg-a spec.replica.promotionToken)
clusters/site-a/db/pooler.yaml, document 0: spec.instances = 0 (site a, Pooler db/pg-a-pooler-rw spec.instances)
clusters/site-b/db/cluster.yaml, document 0: spec.replica.primary = "pg-b" (site b, Cluster db/pg-b spec.replica.primary)
clusters/site-b/db/cluster.yaml, document 0: spec.replica.promotionToken removed (site b, Cluster db/pg-b spec.replica.promotionToken)
clusters/site-b/db/pooler-instances.yaml: spec.instances = 2 (site b, Pooler db/pg-b-pooler-rw spec.instances)
commit message: tack: pg-main primary a to b (planned)
...
Nothing was changed. To run this plan: tack run -f workload.yaml --to b --from a --mode plannedThe tack command lists every command, flag and exit code.
Try it on kind
The end-to-end test builds two kind clusters with Flux, CloudNativePG, Strait and a local Git server, then switches the primary between them. Contributing says what it needs and how to keep the environment to try tack against it.