Otherwise, we'll always get the same sequence of random numbers which
may lead to conflicts. Also bump the number of maximum attempts to 10
instead of 3.
[NO NEW TESTS NEEDED] as I cannot enforce random number collisions.
Existing tests should continue be green and flake slightly less.
Signed-off-by: Valentin Rothberg <rothberg@redhat.com>
Because we cannot reqad the networking mode in the frontent because we
should always use the server default we have to parse the mac and ip
address to the server via a default network. Now when the server reads
the default nsmode it has to reject the provided networks when the mode
is not set to bridge.
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
It is important that we store the current networks from the db in the
config. Also make sure to properly handle aliases and ignore static ip/mac
addresses.
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
Rework the --network parse logic to support multiple networks with
specific network configuration settings.
--network can now be set multiple times. For bridge network mode the
following options have been added:
- **alias=name**: Add network-scoped alias for the container.
- **ip=IPv4**: Specify a static ipv4 address for this container.
- **ip=IPv6**: Specify a static ipv6 address for this container.
- **mac=MAC**: Specify a static mac address address for this container.
- **interface_name**: Specify a name for the created network interface inside the container.
So now you can set --network bridge:ip=10.88.0.10,mac=44:33:22:11:00:99
for the default bridge network as well as for network names.
This is better than using --ip because we can set the ip per network
without any confusion which network the ip address should be assigned
to.
The --ip, --mac-address and --network-alias options are still supported
but --ip or --mac-address can only be set when only one network is set.
This limitation already existed previously.
The ability to specify a custom network interface name is new
Fixes#11534
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
Add the new networks format to specgen. For api users cni_networks is
still supported to make migration easier however the static ip and mac
fields are removed.
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
The swagger api docs used the extra Body struct as part of the request
which is wrong. We just want the plain type.
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
Network connect now supports setting a static ipv4, ipv6 and mac address
for the container network. The options are added to the cli and api.
Fixes#9883
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
Make sure we create new containers in the db with the correct structure.
Also remove some unneeded code for alias handling. We no longer need this
functions.
The specgen format has not been changed for now.
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
The new network db structure stores everything in the networks bucket.
Previously some network settings were not written the the network bucket
and only stored in the container config.
Instead of the old format which used the container ID as value in the
networks buckets we now use the PerNetworkoptions struct there.
To migrate existing users we use the state.GetNetworks() function. If it
fails to read the new format it will automatically migrate the old
config format to the new one. This is allows a flawless migration path.
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
The new network backend stores the networks in a map so the returned
order is not deterministic. Lets sort the network names alphabetically
to ensure a deterministic order.
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
This is the third and hopefully the last attempt to address the flakes
in the pprof tests. We first bumped the timeouts to 2 seconds, then to
5, and since I am running out of ideas let's bump it now to 20 seconds.
Since the timeouts poll, the tests will terminate much earlier but 20
seconds should now really be enough even under highly loaded CI VMs.
Fixes: #12167
Signed-off-by: Valentin Rothberg <rothberg@redhat.com>
... now that they have no public users.
Also remove the HeaderAuthName type, we don't need the type-safety
so much for private constants, and using plain strings results in
less visual noise.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Having a parameter that modifies the provides types.SystemContext
seems rather unexpected and risky to have around - and the only
user of that is actually a no-op, others only provide a nil
SystemContext; so, remove that option and simplify (well, somewhat;
many callers now have extra &types.SystemContext{AuthFilePath}
boilerplate; at least that's consistent with that code carrying
a TODO to create a larger-scope SystemContext).
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Having a parameter that modifies the provides types.SystemContext
seems rather unexpected and risky to have around - and the only
user of that is actually a no-op; so, remove that option and simplify.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
which used to contain more context, but now are just
a pointless copy.
Should not change (test) behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
It is no longer used.
Split the existing tests into MakeXRegistryConfigHeader
and MakeXRegistryAuthHeader variants. For now we don't modify
the implementations at all, to make review simpler; cleanups
will follow.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
... which can be called independently.
For now, there are no new callers, to test that the behavior
has not changed.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
... which can be called independently.
For now, there are no new callers, to test that the behavior
has not changed.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
This shares the code, and makes getConfigCredentials
and getAuthCredentials side-effect free and possibly easier to test.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
... and have GetCredentials pass the values down to
getConfigCredentials and getAuthCredentials.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
It's possibly a bit more expensive, but semantically safer
because it does header normalization.
And we'll regain the cost by not looking up the value repeatedly.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Both have a single caller, so there's no point in looking up
the header value twice.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
In the "no input" case, return a constant instead of
continuing with the decode/convert path, converting empty data.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Don't create a single-element map only for the only caller
to laboriously extract an element of that map; just return
a single entry.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Almost every caller is using it only to wrap an error
in exactly the same way, so move that error context into GetCredentials
and simplify the users.
(The one other caller, build, was even wrapping the error incorrectly
talking about query parameters; so let it use the same text as the others.)
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
... as an end-to-end unit test of the header creation/parsing
code.
Leave the docker.io and docker.io/vendor test cases commented out,
because they are currently failing.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
