Network batching - #631

DeleteInstanceNetworkConfig only calls Bandwidth::Clear when the instance
config actually had a non-zero ingress/egress limit. For unlimited instances
(the common case) nothing was installed, so the previous unconditional Clear
just wasted three tc round-trips (serialized on rtnl_lock) per instance on a
mass teardown. The decision uses the config already held in the network
manager, so no extra state is kept in the bandwidth backend.

Tests to be updated separately.

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

…ilure

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

…stances

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

… fallback

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

Discarding a staged batch without applying it is not expressible with
FlushBatch/Revert: FlushBatch commits and Revert undoes an already
applied batch. Add AbortBatch to both backend interfaces and mocks.

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

FlushBatch commits the firewall, traffic-monitor and storage backends in
sequence and gives each failure its own recovery so a partial batch never
leaves one backend applied while another is not: a firewall-flush failure
aborts the still-staged traffic batch, a traffic-flush failure reverts the
already-flushed firewall, and a storage-commit failure reverts both. The
storage transaction is rolled back on every failure path, and the batch
entries are re-applied per instance so one bad instance is isolated
instead of failing the whole flush.

Cover the firewall-, traffic- and commit-failure paths with tests.

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

ReconcileInstances registered the network DNS server (AdoptDNSServer) only
on the leftover-cleanup path, not for instances adopted as running via
InitInstance. DeleteInstanceNetworkConfig then could not find the DNS
server ("DNS server not found for cleanup") and left stale addnhosts
entries when such an instance was later removed.

Adopt the DNS server before continuing the running-instance branch too.
AdoptDNSServer is idempotent, so multiple instances on one network are
safe; a failure is logged and does not abort adoption.

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>

Adopt a running leftover instance on restart, assert the network DNS
server is registered (CreateServer), and that a subsequent
StopInstanceNetwork drops the instance host entry (RemoveHost). Without
the reconcile fix neither call happens, so the test fails.

Update Start_KeepsLeftoverInstanceWithLiveInterface to expect the DNS
server adoption too.

Signed-off-by: Mykola Solianko <mykola_solianko@epam.com>
Reviewed-by: Mykola Kobets <mykola_kobets@epam.com>
Reviewed-by: Oleksandr Grytsov <oleksandr_grytsov@epam.com>
Reviewed-by: Mykhailo Lohvynenko <mykhailo_lohvynenko@epam.com>