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>
The `RequestPool` method has many args and named returns. This
makes the code hard to follow at times. This commit adds one struct,
`PoolRequest`, to replace these args, and one struct, `AllocatedPool`,
to replace these named returns.
Both structs' fields are properly documented to better define their
semantics, and their relationship with address allocation.
Signed-off-by: Albin Kerouanton <albinker@gmail.com>
UnavailableError is now compatible with errdefs.UnavailableError. These
errors will now return a 503 instead of a 500.
Signed-off-by: Albin Kerouanton <albinker@gmail.com>
InvalidParameter is now compatible with errdefs.InvalidParameter. Thus,
these errors will now return a 400 status code instead of a 500.
Signed-off-by: Albin Kerouanton <albinker@gmail.com>
These errors aren't used in our repo and seem unused by the OSS
community (this was checked with Sourcegraph).
- ErrIpamInternalError has never been used
- ErrInvalidRequest is unused since moby/libnetwork@c85356efa
- ErrPoolNotFound has never been used
- ErrOverlapPool has never been used
- ErrNoAvailablePool has never been used
Signed-off-by: Albin Kerouanton <albinker@gmail.com>
network.requestPoolHelper and Allocator.RequestPool have many args and
output vars with generic types. Add names for them to make it easier to
grasp what's what.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
...that Swarmkit no longer needs now that it has been migrated to use
the new-style driver registration APIs.
Signed-off-by: Cory Snider <csnider@mirantis.com>
The DatastoreConfig discovery type is unused. Remove the constant and
any resulting dead code. Today's biggest loser is the IPAM Allocator:
DatastoreConfig was the only type of discovery event it was listening
for, and there was no other place where a non-nil datastore could be
passed into the allocator. Strip out all the dead persistence code from
Allocator, leaving it as purely an in-memory implementation.
There is no more need to check the consistency of the allocator's
bit-sequences as there is no persistent storage for inconsistent bit
sequences to be loaded from.
Signed-off-by: Cory Snider <csnider@mirantis.com>
The datastore arguments to the IPAM driver Init() functions are always
nil, even in Swarmkit. The only IPAM driver which consumed the
datastores was builtin; all others (null, remote, windowsipam) simply
ignored it. As the signatures of the IPAM driver init functions cannot
be changed without breaking the Swarmkit build, they have to be left
with the same signatures for the time being. Assert that nil datastores
are always passed into the builtin IPAM driver's init function so that
there is no ambiguity the datastores are no longer respected.
Add new Register functions for the IPAM drivers which are free from the
legacy baggage of the Init functions. (The legacy Init functions can be
removed once Swarmkit is migrated to using the Register functions.) As
the remote IPAM driver is the only one which depends on a PluginGetter,
pass it in explicitly as an argument to Register. The other IPAM drivers
should not be forced to depend on a GetPluginGetter() method they do not
use (Interface Segregation Principle).
Signed-off-by: Cory Snider <csnider@mirantis.com>
After moving libnetwork to this repo, we need to update all the import
paths for libnetwork to point to docker/docker/libnetwork instead of
docker/libnetwork.
This change implements that.
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
As part of daemon init, network and ipam drivers are passed a
pluginstore object that implements the plugin/getter interface. Use this
interface methods in libnetwork to interact with network plugins. This
interface provides the new and improved pluginv2 functionality and falls
back to pluginv1 (legacy) if necessary.
Signed-off-by: Anusha Ragunathan <anusha@docker.com>
If a remote plugin returns an empty string in response to RequestAddress(),
the internal helper will return nil which will crash libnetwork in several
places.
Treat an empty string as a new error ipamapi.ErrNoIPReturned.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
- Also restore older behavior where overlap check is not run
when preferred pool is specified. Got broken by recent changes
Signed-off-by: Alessandro Boch <aboch@docker.com>
- So that a DHCP based plugin can express it needs
the endpoint MAC address when requested for an IP address.
- In such case libnetwork will allocate one if not already
provided by user
Signed-off-by: Alessandro Boch <aboch@docker.com>
- Remove from contract predefined errors which are no longer
valid (ex. ErrInvalidIpamService, ErrInvalidIpamConfigService)
- Do not use network driver error for ipam load failure in controller.go
- Bitseq to expose two well-known errors (no more bit available, bit is already set)
- Default ipam to report proper well-known error on RequestAddress()
based on bitseq returned error
- Default ipam errors to comply with types error interface
Signed-off-by: Alessandro Boch <aboch@docker.com>
