Currently endpoint count is maintained as part of
network object and the endpoint count gets updated
frequently while the rest of network is quite stable.
Because of the frequent updates to endpoint count the
network object is getting marshalled and unmarshalled
ferquently. This is causing a lot of churn and transient
memory usage. Fix this by creating a deparate object of
endpoint count so that only that gets updated.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
- Set bridge ipv4 address when bridge is present
- IPv6 changes for bridge
- Convert unit tests to the new model
Signed-off-by: Alessandro Boch <aboch@docker.com>
Currently when docker exits ungracefully it may leave
dangling sandboxes which may hold onto precious network
resources. Added checkpoint state for sandboxes which
on boot up will be used to clean up the sandboxes and
network resources.
On bootup the remaining dangling state in the checkpoint
are read and cleaned up before accepting any new
network allocation requests.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
Always on watching of networks and endpoints can
affect scalability of the cluster beyond a few nodes.
Remove pro active watching and watch only the objects
you are interested in.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
1. Don't save localscope endpoints to localstore for now.
2. Add common function updateToStore/deleteFromStore to store KVObjects.
3. Merge `getNetworksFromGlobalStore` and `getNetworksFromLocalStore`
4. Add `n.isGlobalScoped` before `n.watchEndpoints` in `addNetwork`
5. Fix integration-tests
6. Fix test failure in drivers/remote/driver_test.go
7. Restore network to store if deleteNework failed
Instead of passing the pointer to &ep.iface the current
code is passing the value. So the source variable is not
getting updated properly.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
Currently the endpoint data model consists of multiple
interfaces per-endpoint. This seems to be an overkill
since there is no real use case for it. Removing it
to remove unnecessary complexity from the code.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
- Maps 1 to 1 with container's networking stack
- It holds container's specific nw options which
before were incorrectly owned by Endpoint.
- Sandbox creation no longer coupled with Endpoint Join,
sandbox and endpoint have now separate lifecycle.
- LeaveAll naturally replaced by Sandbox.Delete
- some pkg and file renaming in order to have clear
mapping between structure name and entity ("sandbox")
- Revisited hosts and resolv.conf handling
- Removed from JoinInfo interface capability of setting hosts and resolv.conf paths
- Changed etchosts.Build() to first write the search domains and then the nameservers
Signed-off-by: Alessandro Boch <aboch@docker.com>
- In case of sandboxAdd() failure, drive.Leave() call
in first executed defer reset err to nil. Secondly
executed defer in charge of resetting ep.container to nil
will not get executed.
Signed-off-by: Alessandro Boch <aboch@docker.com>
TL;DR: check for IsExist(err) after a failed MkdirAll() is both
redundant and wrong -- so two reasons to remove it.
Quoting MkdirAll documentation:
> MkdirAll creates a directory named path, along with any necessary
> parents, and returns nil, or else returns an error. If path
> is already a directory, MkdirAll does nothing and returns nil.
This means two things:
1. If a directory to be created already exists, no error is
returned.
2. If the error returned is IsExist (EEXIST), it means there exists
a non-directory with the same name as MkdirAll need to use for
directory. Example: we want to MkdirAll("a/b"), but file "a"
(or "a/b") already exists, so MkdirAll fails.
The above is a theory, based on quoted documentation and my UNIX
knowledge.
3. In practice, though, current MkdirAll implementation [1] returns
ENOTDIR in most of cases described in #2, with the exception when
there is a race between MkdirAll and someone else creating the
last component of MkdirAll argument as a file. In this very case
MkdirAll() will indeed return EEXIST.
Because of #1, IsExist check after MkdirAll is not needed.
Because of #2 and #3, ignoring IsExist error is just plain wrong,
as directory we require is not created. It's cleaner to report
the error now.
[1] https://github.com/golang/go/blob/f9ed2f75/src/os/path.go
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
For the moment in 1.7.1 since we provide a resolv.conf set api
to the driver honor that so that for host driver we can use the
the host's /etc/resolv.conf file as is rather than putting the
contents through a filtering logic.
It should be noted that the driver side capability to set the
resolv.conf file is most likely going to go away in the future
but this should be fine for 1.7.1
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
In that commit, AtomicPutCreate takes previous = nil to Atomically create keys
that don't exist. We need a create operation that is atomic to prevent races
between multiple libnetworks creating the same object.
Previously, we just created new KVs with an index of 0 and wrote them to the
datastore. Consul accepts this behaviour and interprets index of 0 as
non-existing, but other data backends do no.
- Add Exists() to the KV interface. SetIndex() should also modify a KV so
that it exists.
- Call SetIndex() from within the GetObject() method on DataStore interface.
- This ensures objects have the updated values for exists and index.
- Add SetValue() to the KV interface. This allows implementers to define
their own method to marshall and unmarshall (as bitseq and allocator have).
- Update existing users of the DataStore (endpoint, network, bitseq,
allocator, ov_network) to new interfaces.
- Fix UTs.
Currently container can join one endpoint when it is started.
More endpoints can be attached at a later point in time. But
when that happens this attachment should only have meaning
only as long as the container is alive. The attachment should
lose it's meaning when the container goes away. Cuurently there
is no way for the container management code to tell libnetwork
to detach the container from all attached endpoints. This PR
provides an additional API `LeaveAll` which adds this
functionality,
To facilitate this and make the sanbox lifecycle consistent
some slight changes have been made to the behavior of sandbox
management code. The sandbox is no longer destroyed when the
last endpoint is detached from the container. Instead the sandbox
ie kept alive and can only be destroyed with a `LeaveAll` call.
This gives better control of sandbox lifecycle by the container
management code and the sandbox doesn't get destroyed from under
the carpet while the container is still using it.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
Add a minimal service discover support using service names or
service names qualified with network name. This is achieved
by populating the container's /etc/hosts file record with the
appropriate entries
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
* Removed network from being marshalled (it is part of the key anyways)
* Reworked the watch function to handle container-id on endpoints
* Included ContainerInfo to be marshalled which needs to be synchronized
* Resolved multiple race issues by introducing data locks
Signed-off-by: Madhu Venugopal <madhu@docker.com>
Unless it is a forbidden error, libnetwork should not fail a forced
delete of a network and endpoint if the driver throws an error.
Signed-off-by: Madhu Venugopal <madhu@docker.com>
When an endpoint is joined by a container it may
optionally pass a priority to resolve resource
conflicts inside the sandbox when more than one
endpoint provides the same kind of resource. If the
the priority is the same for two endpoints with
conflicting resources then the endpoint network names
are used to resolve the conflict.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
Only remove the interfaces owned by the endpoint from
the sandbox when the container leaves the endpoint.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
Now that Endpoint interface has the Info method there is no
need to return container data as a return value in the Join
method. Removed the return value and fixed all the callers.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
Currently the driver api allows the driver to specify the
full interface name for the interface inside the container.
This is not appropriate since the driver does not have the full
view of the sandbox to correcly allocate an unambiguous interface
name. Instead with this PR the driver will be allowed to specify
a prefix for the name and libnetwork and sandbox layers will
disambiguate it with an appropriate suffix.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
- Package types to define the interfaces libnetwork errors
may implement, so that caller can categorize them.
Signed-off-by: Alessandro Boch <aboch@docker.com>
This is need to decouple types from netutils which has linux
dependencies. This way the client code which needs network types
can just pull in types package which makes client code platform
agnostic.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
The container's /etc/resolv.conf permission was getting setup
as 0600 while it should be 0644 for every user inside the
container to be able to read it. The tempfile that we create
initially to populate the resolvconf content is getting created
with 0600 mode. Changed it to 0644 once it is created since there
is noway to pass mode option to ioutil.Tempfile
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
Refactored the driver api so that is aligns well with the design
of endpoint lifecycle becoming decoupled from the container lifecycle.
Introduced go interfaces to obtain address information during CreateEndpoint.
Go interfaces are also used to get data from driver during join.
This sort of deisgn hides the libnetwork specific type details from drivers.
Another adjustment is to provide a list of interfaces during CreateEndpoint. The
goal of this is many-fold:
* To indicate to the driver that IP address has been assigned by some other
entity (like a user wanting to use their own static IP for an endpoint/container)
and asking the driver to honor this. Driver may reject this configuration
and return an error but it may not try to allocate an IP address and override
the passed one.
* To indicate to the driver that IP address has already been allocated once
for this endpoint by an instance of the same driver in some docker host
in the cluster and this is merely a notification about that endpoint and the
allocated resources.
* In case the list of interfaces is empty the driver is required to allocate and
assign IP addresses for this endpoint.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
If resolvConfPath is unavailable and if the internally generated .hash file
is still present, then updateDNS should not consider the presence of internally
generated .hash. Rather, it must handle it as a case of using this
resolvConfPath for the first time.
Signed-off-by: Madhu Venugopal <madhu@docker.com>
Jana Radhakrishnan