isle/go/daemon/children/children.go

// Package children manages the creation, lifetime, and shutdown of child
// processes created by the daemon.
package children

import (
	"context"
	"errors"
	"fmt"

	"code.betamike.com/micropelago/pmux/pmuxlib"
	"dev.mediocregopher.com/mediocre-go-lib.git/mctx"
	"dev.mediocregopher.com/mediocre-go-lib.git/mlog"

	"isle/bootstrap"
	"isle/daemon/daecommon"
	"isle/secrets"
	"isle/toolkit"
)

// Opts are optional parameters which can be passed in when initializing a new
// Children instance. A nil Opts is equivalent to a zero value.
type Opts struct{}

func (o *Opts) withDefaults() *Opts {
	if o == nil {
		o = new(Opts)
	}

	return o
}

// Children manages all child processes of a network. Child processes are
// comprised of:
// - nebula
// - dnsmasq
// - garage (0 or more, depending on configured storage allocations)
type Children struct {
	logger           *mlog.Logger
	binDirPath       string
	runtimeDir       toolkit.Dir
	garageAdminToken string
	opts             Opts

	garageRPCSecret string

	nebulaProc  *pmuxlib.Process
	dnsmasqProc *pmuxlib.Process
	garageProcs map[string]*pmuxlib.Process
}

// New initializes and returns a Children instance. If initialization fails an
// error is returned.
func New(
	ctx context.Context,
	logger *mlog.Logger,
	binDirPath string,
	secretsStore secrets.Store,
	networkConfig daecommon.NetworkConfig,
	runtimeDir toolkit.Dir,
	garageAdminToken string,
	hostBootstrap bootstrap.Bootstrap,
	opts *Opts,
) (
	*Children, error,
) {
	opts = opts.withDefaults()

	logger.Info(ctx, "Loading secrets")
	garageRPCSecret, err := daecommon.GetGarageRPCSecret(ctx, secretsStore)
	if err != nil && !errors.Is(err, secrets.ErrNotFound) {
		return nil, fmt.Errorf("loading garage RPC secret: %w", err)
	}

	c := &Children{
		logger:           logger,
		binDirPath:       binDirPath,
		runtimeDir:       runtimeDir,
		garageAdminToken: garageAdminToken,
		opts:             *opts,
		garageRPCSecret:  garageRPCSecret,
	}

	if c.nebulaProc, err = nebulaPmuxProc(
		ctx,
		c.logger,
		c.runtimeDir.Path,
		c.binDirPath,
		networkConfig,
		hostBootstrap,
	); err != nil {
		return nil, fmt.Errorf("starting nebula: %w", err)
	}

	if err := waitForNebula(ctx, c.logger, hostBootstrap); err != nil {
		logger.Warn(ctx, "Failed waiting for nebula to initialize, shutting down child processes", err)
		c.Shutdown()
		return nil, fmt.Errorf("waiting for nebula to start: %w", err)
	}

	if c.dnsmasqProc, err = dnsmasqPmuxProc(
		ctx,
		c.logger,
		c.runtimeDir.Path,
		c.binDirPath,
		networkConfig,
		hostBootstrap,
	); err != nil {
		logger.Warn(ctx, "Failed to start dnsmasq, shutting down child processes", err)
		c.Shutdown()
		return nil, fmt.Errorf("starting dnsmasq: %w", err)
	}

	// TODO wait for dnsmasq to come up

	if c.garageProcs, err = garagePmuxProcs(
		ctx,
		c.logger,
		garageRPCSecret,
		c.runtimeDir.Path,
		c.binDirPath,
		networkConfig,
		garageAdminToken,
		hostBootstrap,
	); err != nil {
		logger.Warn(ctx, "Failed to start garage processes, shutting down child processes", err)
		c.Shutdown()
		return nil, fmt.Errorf("starting garage processes: %w", err)
	}

	if err := waitForGarage(
		ctx, c.logger, networkConfig, garageAdminToken, hostBootstrap,
	); err != nil {
		logger.Warn(ctx, "Failed waiting for garage processes to initialize, shutting down child processes", err)
		c.Shutdown()
		return nil, fmt.Errorf("waiting for garage processes to initialize: %w", err)
	}

	return c, nil
}

// TODO block until process has been confirmed to have come back up
// successfully.
func (c *Children) reloadDNSMasq(
	ctx context.Context,
	networkConfig daecommon.NetworkConfig,
	hostBootstrap bootstrap.Bootstrap,
) error {
	if _, changed, err := dnsmasqWriteConfig(
		ctx, c.logger, c.runtimeDir.Path, networkConfig, hostBootstrap,
	); err != nil {
		return fmt.Errorf("writing new dnsmasq config: %w", err)
	} else if !changed {
		c.logger.Info(ctx, "No changes to dnsmasq config file")
		return nil
	}

	c.logger.Info(ctx, "dnsmasq config file has changed, restarting process")
	c.dnsmasqProc.Restart()

	return nil
}

func (c *Children) reloadNebula(
	ctx context.Context,
	networkConfig daecommon.NetworkConfig,
	hostBootstrap bootstrap.Bootstrap,
) error {
	if _, changed, err := nebulaWriteConfig(
		ctx, c.logger, c.runtimeDir.Path, networkConfig, hostBootstrap,
	); err != nil {
		return fmt.Errorf("writing a new nebula config: %w", err)
	} else if !changed {
		c.logger.Info(ctx, "No changes to nebula config file")
		return nil
	}

	c.logger.Info(ctx, "nebula config file has changed, restarting process")
	c.nebulaProc.Restart()

	if err := waitForNebula(ctx, c.logger, hostBootstrap); err != nil {
		return fmt.Errorf("waiting for nebula to start: %w", err)
	}

	return nil
}

func (c *Children) reloadGarage(
	ctx context.Context,
	networkConfig daecommon.NetworkConfig,
	hostBootstrap bootstrap.Bootstrap,
) error {
	allocs := networkConfig.Storage.Allocations
	if len(allocs) == 0 {
		return nil
	}

	var anyChanged bool
	for _, alloc := range allocs {
		var (
			procName = garagePmuxProcName(alloc)
			ctx      = mctx.Annotate(
				ctx,
				"garageProcName", procName,
				"garageDataPath", alloc.DataPath,
			)
		)

		// TODO it's possible that the config changed, but only the bootstrap
		// peers, in which case we don't need to restart the node.
		childConfigPath, changed, err := garageWriteChildConfig(
			ctx,
			c.logger,
			c.garageRPCSecret,
			c.runtimeDir.Path,
			c.garageAdminToken,
			hostBootstrap,
			alloc,
		)
		if err != nil {
			return fmt.Errorf("writing child config file for alloc %+v: %w", alloc, err)
		} else if !changed {
			c.logger.Info(ctx, "No changes to garage config file")
			continue
		}

		anyChanged = true

		if proc, ok := c.garageProcs[procName]; ok {
			c.logger.Info(ctx, "garage config has changed, restarting process")
			proc.Restart()
			continue
		}

		c.logger.Info(ctx, "garage config has been added, creating process")
		c.garageProcs[procName] = garagePmuxProc(
			ctx, c.logger, c.binDirPath, procName, childConfigPath,
		)
	}

	if anyChanged {
		if err := waitForGarage(
			ctx, c.logger, networkConfig, c.garageAdminToken, hostBootstrap,
		); err != nil {
			return fmt.Errorf("waiting for garage to start: %w", err)
		}
	}

	return nil
}

// Reload applies a ReloadDiff to the Children, using the given bootstrap which
// must be the same one which was passed to CalculateReloadDiff.
func (c *Children) Reload(
	ctx context.Context,
	newNetworkConfig daecommon.NetworkConfig,
	newBootstrap bootstrap.Bootstrap,
) error {
	if err := c.reloadNebula(ctx, newNetworkConfig, newBootstrap); err != nil {
		return fmt.Errorf("reloading nebula: %w", err)
	}

	var errs []error

	if err := c.reloadDNSMasq(ctx, newNetworkConfig, newBootstrap); err != nil {
		errs = append(errs, fmt.Errorf("reloading dnsmasq: %w", err))
	}

	if err := c.reloadGarage(ctx, newNetworkConfig, newBootstrap); err != nil {
		errs = append(errs, fmt.Errorf("reloading garage: %w", err))
	}

	return errors.Join(errs...)
}

// Shutdown blocks until all child processes have gracefully shut themselves
// down.
func (c *Children) Shutdown() {
	for _, proc := range c.garageProcs {
		proc.Stop()
	}

	if c.dnsmasqProc != nil {
		c.dnsmasqProc.Stop()
	}

	if c.nebulaProc != nil {
		c.nebulaProc.Stop()
	}
}