mirror of
https://github.com/moby/moby.git
synced 2026-08-08 00:52:17 +00:00
The previous allocator was subnetting address pools eagerly when the daemon started, and would then just iterate over that list whenever RequestPool was called. This was leading to high memory usage whenever IPv6 pools were configured with a target subnet size too different from the pools prefix size. For instance: pool = fd00::/8, target size = /64 -- 2 ^ (64-8) subnets would be generated upfront. This would take approx. 9 * 10^18 bits -- way too much for any human computer in 2024. Another noteworthy issue, the previous implementation was allocating a subnet, and then in another layer was checking whether the allocation was conflicting with some 'reserved networks'. If so, the allocation would be retried, etc... To make it worse, 'reserved networks' would be recomputed on every iteration. This is totally ineffective as there could be 'reserved networks' that fully overlap a given address pool (or many!). To fix this issue, a new field `Exclude` is added to `RequestPool`. It's up to each driver to take it into account. Since we don't know whether this retry loop is useful for some remote IPAM driver, it's reimplemented bug-for-bug directly in the remote driver. The new allocator uses a linear-search algorithm. It takes advantage of all lists (predefined pools, allocated subnets and reserved networks) being sorted and logically combines 'allocated' and 'reserved' through a 'double cursor' to iterate on both lists at the same time while preserving the total order. At the same time, it iterates over 'predefined' pools and looks for the first empty space that would be a good fit. Currently, the size of the allocated subnet is still dictated by each 'predefined' pools. We should consider hardcoding that size instead, and let users specify what subnet size they want. This wasn't possible before as the subnets were generated upfront. This new allocator should be able to deal with this easily. The method used for static allocation has been updated to make sure the ascending order of 'allocated' is preserved. It's bug-for-bug compatible with the previous implementation. One consequence of this new algorithm is that we don't keep track of where the last allocation happened, we just allocate the first free subnet we find. Before: - Allocate: 10.0.1.0/24, 10.0.2.0/24 ; Deallocate: 10.0.1.0/24 ; Allocate 10.0.3.0/24. Now, the 3rd allocation would yield 10.0.1.0/24 once again. As it doesn't change the semantics of the allocator, there's no reason to worry about that. Finally, about 'reserved networks'. The heuristics we use are now properly documented. It was discovered that we don't check routes for IPv6 allocations -- this can't be changed because there's no such thing as on-link routes for IPv6. (Kudos to Rob Murray for coming up with the linear-search idea.) Signed-off-by: Albin Kerouanton <albinker@gmail.com>
256 lines
8.0 KiB
Go
256 lines
8.0 KiB
Go
package remote
|
|
|
|
import (
|
|
"context"
|
|
"fmt"
|
|
"net"
|
|
"net/netip"
|
|
|
|
"github.com/containerd/log"
|
|
"github.com/docker/docker/libnetwork/ipamapi"
|
|
"github.com/docker/docker/libnetwork/ipams/remote/api"
|
|
"github.com/docker/docker/libnetwork/types"
|
|
"github.com/docker/docker/pkg/plugingetter"
|
|
"github.com/docker/docker/pkg/plugins"
|
|
"github.com/pkg/errors"
|
|
)
|
|
|
|
type allocator struct {
|
|
endpoint *plugins.Client
|
|
name string
|
|
}
|
|
|
|
// PluginResponse is the interface for the plugin request responses
|
|
type PluginResponse interface {
|
|
IsSuccess() bool
|
|
GetError() string
|
|
}
|
|
|
|
func newAllocator(name string, client *plugins.Client) ipamapi.Ipam {
|
|
a := &allocator{name: name, endpoint: client}
|
|
return a
|
|
}
|
|
|
|
// Register registers a remote ipam when its plugin is activated.
|
|
func Register(cb ipamapi.Registerer, pg plugingetter.PluginGetter) error {
|
|
newPluginHandler := func(name string, client *plugins.Client) {
|
|
a := newAllocator(name, client)
|
|
if cps, err := a.(*allocator).getCapabilities(); err == nil {
|
|
if err := cb.RegisterIpamDriverWithCapabilities(name, a, cps); err != nil {
|
|
log.G(context.TODO()).Errorf("error registering remote ipam driver %s due to %v", name, err)
|
|
}
|
|
} else {
|
|
log.G(context.TODO()).Infof("remote ipam driver %s does not support capabilities", name)
|
|
log.G(context.TODO()).Debug(err)
|
|
if err := cb.RegisterIpamDriver(name, a); err != nil {
|
|
log.G(context.TODO()).Errorf("error registering remote ipam driver %s due to %v", name, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Unit test code is unaware of a true PluginStore. So we fall back to v1 plugins.
|
|
handleFunc := plugins.Handle
|
|
if pg != nil {
|
|
handleFunc = pg.Handle
|
|
activePlugins := pg.GetAllManagedPluginsByCap(ipamapi.PluginEndpointType)
|
|
for _, ap := range activePlugins {
|
|
client, err := getPluginClient(ap)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
newPluginHandler(ap.Name(), client)
|
|
}
|
|
}
|
|
handleFunc(ipamapi.PluginEndpointType, newPluginHandler)
|
|
return nil
|
|
}
|
|
|
|
func getPluginClient(p plugingetter.CompatPlugin) (*plugins.Client, error) {
|
|
if v1, ok := p.(plugingetter.PluginWithV1Client); ok {
|
|
return v1.Client(), nil
|
|
}
|
|
|
|
pa, ok := p.(plugingetter.PluginAddr)
|
|
if !ok {
|
|
return nil, errors.Errorf("unknown plugin type %T", p)
|
|
}
|
|
|
|
if pa.Protocol() != plugins.ProtocolSchemeHTTPV1 {
|
|
return nil, errors.Errorf("unsupported plugin protocol %s", pa.Protocol())
|
|
}
|
|
|
|
addr := pa.Addr()
|
|
client, err := plugins.NewClientWithTimeout(addr.Network()+"://"+addr.String(), nil, pa.Timeout())
|
|
if err != nil {
|
|
return nil, errors.Wrap(err, "error creating plugin client")
|
|
}
|
|
return client, nil
|
|
}
|
|
|
|
func (a *allocator) call(methodName string, arg interface{}, retVal PluginResponse) error {
|
|
method := ipamapi.PluginEndpointType + "." + methodName
|
|
err := a.endpoint.Call(method, arg, retVal)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if !retVal.IsSuccess() {
|
|
return fmt.Errorf("remote: %s", retVal.GetError())
|
|
}
|
|
return nil
|
|
}
|
|
|
|
func (a *allocator) getCapabilities() (*ipamapi.Capability, error) {
|
|
var res api.GetCapabilityResponse
|
|
if err := a.call("GetCapabilities", nil, &res); err != nil {
|
|
return nil, err
|
|
}
|
|
return res.ToCapability(), nil
|
|
}
|
|
|
|
// GetDefaultAddressSpaces returns the local and global default address spaces
|
|
func (a *allocator) GetDefaultAddressSpaces() (string, string, error) {
|
|
res := &api.GetAddressSpacesResponse{}
|
|
if err := a.call("GetDefaultAddressSpaces", nil, res); err != nil {
|
|
return "", "", err
|
|
}
|
|
return res.LocalDefaultAddressSpace, res.GlobalDefaultAddressSpace, nil
|
|
}
|
|
|
|
// RequestPool requests an address pool in the specified address space.
|
|
//
|
|
// This is a bug-for-bug re-implementation of the logic originally found in
|
|
// requestPoolHelper prior to v27. See https://github.com/moby/moby/blob/faf84d7f0a1f2e6badff6f720a3e1e559c356fff/libnetwork/network.go#L1518-L1570
|
|
func (a *allocator) RequestPool(req ipamapi.PoolRequest) (ipamapi.AllocatedPool, error) {
|
|
var tmpPoolLeases []string
|
|
defer func() {
|
|
// Release all pools we held on to.
|
|
for _, pID := range tmpPoolLeases {
|
|
if err := a.ReleasePool(pID); err != nil {
|
|
log.G(context.TODO()).Warnf("Failed to release overlapping pool")
|
|
}
|
|
}
|
|
}()
|
|
|
|
_, globalSpace, err := a.GetDefaultAddressSpaces()
|
|
if err != nil {
|
|
return ipamapi.AllocatedPool{}, err
|
|
}
|
|
|
|
remoteReq := &api.RequestPoolRequest{
|
|
AddressSpace: req.AddressSpace,
|
|
Pool: req.Pool,
|
|
SubPool: req.SubPool,
|
|
Options: req.Options,
|
|
V6: req.V6,
|
|
}
|
|
|
|
for {
|
|
alloc, err := a.requestPool(remoteReq)
|
|
if err != nil {
|
|
return alloc, err
|
|
}
|
|
|
|
// If the network pool was explicitly chosen, the network belongs to
|
|
// global address space, or it is invalid ("0.0.0.0/0"), then we don't
|
|
// perform check for overlaps.
|
|
//
|
|
// FIXME(thaJeztah): why are we ignoring invalid pools here?
|
|
//
|
|
// The "invalid" conditions was added in [libnetwork#1095][1], which
|
|
// moved code to reduce os-specific dependencies in the ipam package,
|
|
// but also introduced a types.IsIPNetValid() function, which considers
|
|
// "0.0.0.0/0" invalid, and added it to the conditions below.
|
|
//
|
|
// Unfortunately review does not mention this change, so there's no
|
|
// context why. Possibly this was done to prevent errors further down
|
|
// the line (when checking for overlaps), but returning an error here
|
|
// instead would likely have avoided that as well, so we can only guess.
|
|
//
|
|
// [1]: https://github.com/moby/libnetwork/commit/5ca79d6b87873264516323a7b76f0af7d0298492#diff-bdcd879439d041827d334846f9aba01de6e3683ed8fdd01e63917dae6df23846
|
|
if req.Pool != "" || req.AddressSpace == globalSpace || alloc.Pool.String() == "0.0.0.0/0" {
|
|
return alloc, nil
|
|
}
|
|
|
|
// Check for overlap and if none found, we have found the right pool.
|
|
if !checkOverlaps(alloc, req.Exclude) {
|
|
return alloc, nil
|
|
}
|
|
|
|
// Pool obtained in this iteration is overlapping. Hold onto the pool
|
|
// and don't release it yet, because we don't want IPAM to give us back
|
|
// the same pool over again. But make sure we still do a deferred release
|
|
// when we have either obtained a non-overlapping pool or ran out of
|
|
// pre-defined pools.
|
|
tmpPoolLeases = append(tmpPoolLeases, alloc.PoolID)
|
|
}
|
|
}
|
|
|
|
func (a *allocator) requestPool(req *api.RequestPoolRequest) (ipamapi.AllocatedPool, error) {
|
|
res := &api.RequestPoolResponse{}
|
|
if err := a.call("RequestPool", req, res); err != nil {
|
|
return ipamapi.AllocatedPool{}, err
|
|
}
|
|
|
|
retPool, err := netip.ParsePrefix(res.Pool)
|
|
return ipamapi.AllocatedPool{
|
|
PoolID: res.PoolID,
|
|
Pool: retPool,
|
|
Meta: res.Data,
|
|
}, err
|
|
}
|
|
|
|
// checkOverlaps returns true if the 'pool' overlaps with some prefix in 'reserved'.
|
|
func checkOverlaps(pool ipamapi.AllocatedPool, reserved []netip.Prefix) bool {
|
|
for _, r := range reserved {
|
|
if r.Overlaps(pool.Pool) {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
}
|
|
|
|
// ReleasePool removes an address pool from the specified address space
|
|
func (a *allocator) ReleasePool(poolID string) error {
|
|
req := &api.ReleasePoolRequest{PoolID: poolID}
|
|
res := &api.ReleasePoolResponse{}
|
|
return a.call("ReleasePool", req, res)
|
|
}
|
|
|
|
// RequestAddress requests an address from the address pool
|
|
func (a *allocator) RequestAddress(poolID string, address net.IP, options map[string]string) (*net.IPNet, map[string]string, error) {
|
|
var (
|
|
prefAddress string
|
|
retAddress *net.IPNet
|
|
err error
|
|
)
|
|
if address != nil {
|
|
prefAddress = address.String()
|
|
}
|
|
req := &api.RequestAddressRequest{PoolID: poolID, Address: prefAddress, Options: options}
|
|
res := &api.RequestAddressResponse{}
|
|
if err := a.call("RequestAddress", req, res); err != nil {
|
|
return nil, nil, err
|
|
}
|
|
if res.Address != "" {
|
|
retAddress, err = types.ParseCIDR(res.Address)
|
|
} else {
|
|
return nil, nil, ipamapi.ErrNoIPReturned
|
|
}
|
|
return retAddress, res.Data, err
|
|
}
|
|
|
|
// ReleaseAddress releases the address from the specified address pool
|
|
func (a *allocator) ReleaseAddress(poolID string, address net.IP) error {
|
|
var relAddress string
|
|
if address != nil {
|
|
relAddress = address.String()
|
|
}
|
|
req := &api.ReleaseAddressRequest{PoolID: poolID, Address: relAddress}
|
|
res := &api.ReleaseAddressResponse{}
|
|
return a.call("ReleaseAddress", req, res)
|
|
}
|
|
|
|
func (a *allocator) IsBuiltIn() bool {
|
|
return false
|
|
}
|