Documentation
¶
Overview ¶
Package netpool hands out pre-created swarm overlays.
A swarm service cannot join a network without rolling its tasks, so traefik cannot follow environments that come and go: it would roll on every new env and every PR preview. Instead the networks exist up front, traefik joins all of them once, and an environment takes a free one and gives it back (docs/plans/30-docker-swarm.md, decision 2).
The consequence is that a network name means nothing on its own. The only mapping from network to owner is the DB column, environments.network for an env, tiles.shared_net for a shared managed instance. Nothing parses a name.
Index ¶
- Constants
- func Backfill(ctx context.Context, store repo.Store, rt *runtime.Runtime)
- func ClaimDB(ctx context.Context, store repo.Store, rt *runtime.Runtime, t *repo.Tile) (string, error)
- func ClaimEnv(ctx context.Context, store repo.Store, rt *runtime.Runtime, envID string) (string, error)
- func Ensure(ctx context.Context, rt *runtime.Runtime) error
- func ReleaseDB(ctx context.Context, store repo.Store, rt *runtime.Runtime, t *repo.Tile) error
- func ReleaseEnv(ctx context.Context, store repo.Store, rt *runtime.Runtime, ...) error
- func Sweep(ctx context.Context, store repo.Store, rt *runtime.Runtime)
Constants ¶
const ( // EnvPrefix names the environment pool, DBPrefix the pool a shared // managed db draws from so consumers in dev cannot see consumers in prod. EnvPrefix = "stkr-net-" DBPrefix = "stkr-dbnet-" // Sizes pre-created at boot. Not a ceiling: a claim past the end creates // the next name, which is all growth means while traefik is still a // container that can hot-join. Once traefik is a service (step 4) growth // has to roll it, and that is where the warning on the plan page lands. EnvPoolSize = 32 DBPoolSize = 8 )
Variables ¶
This section is empty.
Functions ¶
func Backfill ¶
Sweep clears out pooled networks that no row holds any more. Shaped like runtime.SweepForwardState: a crash between deleting the owner row and disconnecting its containers leaves a network that is free by the DB and occupied in reality, so the next environment to claim it starts with a neighbour it can reach. Best-effort, boot only. Backfill gives an overlay to every environment that has none.
Migration 009 added environments.network with an empty default and backfilled nothing, and the only thing that ever fills it is ClaimEnv, which runs at deploy. So on an upgraded install every environment that already existed sat on no overlay until somebody happened to redeploy it: port forward refused with "tile is not on a network yet", and the proxy left the whole environment out of its network-to-env map without saying anything.
Runs at boot next to Sweep. A pool with nothing left is logged and skipped rather than fatal, an install that cannot claim is still an install that should come up and say so.
func ClaimDB ¶
func ClaimDB(ctx context.Context, store repo.Store, rt *runtime.Runtime, t *repo.Tile) (string, error)
ClaimDB is ClaimEnv for a shared managed instance's own network.
func ClaimEnv ¶
func ClaimEnv(ctx context.Context, store repo.Store, rt *runtime.Runtime, envID string) (string, error)
ClaimEnv returns the environment's overlay, taking a free one from the pool the first time. Idempotent: an env that already holds one keeps it.
Types ¶
This section is empty.