Robert Kukura [Mon, 10 Mar 2014 19:06:09 +0000 (15:06 -0400)]
ML2: Remove validate_port_binding() and unbind_port()
The API implemented by ML2 mechanism drivers included three methods
related to port binding that were called within DB transactions, but
that could potentially involve communication with controllers or
devices that should not be done within transactions. A subsequent
patch will move the calls to bind_port() outside of tranactions. This
patch eliminates the other two methods from the MechanismDriver API.
The validate_port_binding() method was previously called on the bound
mechanism driver to check whether an existing binding was still valid,
so that the port could be rebound if something changed. But since nova
has no way to handle changes to binding:vif_type or
binding:vif_details after a port is initially plugged, this turned out
not to be useful, so the method has been removed from the
MechanismDriver API. Now, once a port is successfully bound, the
binding remains until the port is deleted or any of it's
binding:host_id, binding:vnic_type, or binding:profile attribute
values are changed.
The unbind_port() method was previously called on the bound mechanism
driver as an existing binding was removed. This method was not used by
any existing mechanism drivers, and was redundant with the
update_port_precommit() and update_port_postcommit() methods that are
called on all mechanism drivers when an existing binding is removed,
so this method has also been removed from the driver API.
Eliminating the unbind_port() call allows the binding details to be
made available via the PortContext in delete_port_postcommit() calls,
completing the resolution of bug 1276395.
This patch adds a semaphore for serializing the delete port API
calls and update_port_status calls to avoid simultaneous
attempts to acquire SQL update locks on the port table from
the same thread pool.
This semaphore has been introduced to avoid undesired eventlet
yields which trigger 'lock wait timeout' errors in the DB backend;
it should have a negligible impact on the overall performance.
Mark McClain [Fri, 7 Mar 2014 20:07:43 +0000 (15:07 -0500)]
add HEAD sentinel file that contains migration revision
This change wraps the command to generate a migration with code to
update a file called HEAD. The HEAD file will contain the revision ID
of the head of the migraton timeline. Additionally, check migrations has
been altered to verify the contents of this file against the timeline
head.
This file helps the OpenStack gate detect potential migration branches
without running alembic via git merge conflicts.
Akihiro Motoki [Wed, 19 Mar 2014 08:08:37 +0000 (17:08 +0900)]
NEC plugin: Honor Retry-After response from OFC
A backend OpenFlow controller nec plugin talks to can return
503 response with retry-after header when it is busy.
It is better to honor retry-after header to avoid unnecessary
user-visible errors due to temporary busy condition.
Irena Berezovsky [Tue, 18 Mar 2014 14:35:35 +0000 (16:35 +0200)]
Add update binding:profile with physical_network
Add temporary solution in order to support multiple physical networks
by mlnx ML2 MechanismDriver.
Due to non merged patches in nova that should support propagating
physical_network retrieved from port binding:profile attribute
to VIF/Network object.
The code will be removed once relevant nova patches are merged.
The code is disabled by default and should be enabled via
ml2_conf_mlnx.ini config file.
Itsuro Oda [Tue, 17 Dec 2013 01:16:51 +0000 (10:16 +0900)]
Enable to select an RPC handling plugin under Metaplugin
Multiple plugins under metaplugin become 'q-plugin' topic
consumers and a request from an agent is handled by one of
them randomly. Fortunatly most of RPC callbacks are common
for plugins but a problem occurs if an RPC is not supported
by the received plugin.
This is one of risks when using metaplugin. Fundamental fix
of this problem (such as RPC delegation handling of metaplugin)
is difficult since each plugin needs to modify.
But when only one plugin has plugin specific RPCs and other
RPCs are independet of plugins, if the plugin can be selected
for RPC handling, the problem does not happen. Typical use
case of metaplugin such as combination of an agent-based
plugin and a controller-based plugin often applies to this
condition.
This patch adds 'rpc_flavor' configuration parameter to
select an RPC handling plugin. If 'rpc_flavor' is specified,
only the specified plugin becomes 'q-plugin' topic consumer.
If 'rpc_flavor' is not specified, the behavior is same as
previous one.
Kevin Benton [Tue, 4 Mar 2014 11:27:11 +0000 (03:27 -0800)]
Mock agent RPC for FWaaS tests to delete DB objs
This changes the firewall service unit tests to
mock the RPC calls from the firewall service to
the agent. This allows the tests to fake the
agent response RPC that removes the firewall
from the DB so the all of the other objects that
the firewall depends on can be deleted.
Kevin Benton [Tue, 18 Mar 2014 00:06:46 +0000 (17:06 -0700)]
Allow CIDRs with non-zero masked portions
Allow users to specify CIDRs with bits other
than zeros in the masked portion of the subnet.
e.g. 192.168.1.5/24 is accepted and converted
to 192.168.1.0/24.
Dane LeBlanc [Tue, 18 Mar 2014 16:37:33 +0000 (12:37 -0400)]
Cisco plugin fails with ParseError no elem found
When the Cisco nexus plugin is configured on DevStack, the tempest
Neutron API test test_port_list_filter_by_router_id fails with the
following error:
ParseError: no element found: line 1, column 0
The root cause is that the Cisco Nexus plugin model layer
is encapsulating each dictionary response from the OVS subplugin
into a list, whereas the WSGI layer is expecting a dictionary
response. The fix is not to encapsulate these responses into a list.
Kevin Benton [Tue, 18 Mar 2014 19:11:53 +0000 (12:11 -0700)]
ML2 BigSwitch: Don't modify parent context
Makes a copy of the port context before changing
it in preparation for the backend controller so
other drivers are not affected. Also removes a
UT that was exercising direct modification of VIF
details in the port context, which is not allowed
by the ML2 plugin.
Assaf Muller [Sun, 16 Mar 2014 11:01:18 +0000 (13:01 +0200)]
Change report_interval from 4 to 30, agent_down_time from 9 to 75
report_interval is how often an agent sends out a heartbeat to the
service. The Neutron service responds to these 'report_state' RPC
messages by updating the agent's heartbeat DB record.
The last heartbeat is then compared to the configured
agent_down_time to determine if the agent is up or down.
The agent's status is used when scheduling networks on DHCP
and L3 agents.
In the spirit of sane defaults suited for production, these values
should be bumped to reduce the load on the Neutron service
dramatically, freeing up CPU time to perform intensive operations.
Maru Newby [Fri, 14 Mar 2014 22:14:09 +0000 (22:14 +0000)]
Stop removing ip allocations on port delete
The _delete_port() method was manually removing related
IPAllocation instances despite the existence of a perfectly
good cascade deletion relationship in the model. This patch
puts an end to that nonsense and the potential for deadlock that
it represented.
Aaron Rosen [Thu, 13 Mar 2014 05:22:53 +0000 (22:22 -0700)]
NSX: Fix router-interface-delete returns 404 when router not in nsx
Previously, if one would run router-interface-delete and the router
was not found in NSX. Neutron would raise a 404 error and remove the interface
from the database. In this case it would be better if we did not raise
a 404 error as the caller shouldn't be aware that the backend is out of sync.
Dazhao [Wed, 13 Nov 2013 10:24:32 +0000 (18:24 +0800)]
Calculate stateless IPv6 address
In order to support stateless IPv6, neutron should calculate IPv6 address base
on IPv6 prefix and MAC address via EUI-64 specification. This change adds
common library to calculate IPv6 address.
Partially-implements bp ipv6-two-attributes Co-Authored-By: Xu Han Peng <xuhanp@cn.ibm.com>
Change-Id: I68ccbc42388ec760d6fead242e080822ca753913
Sean M. Collins [Thu, 30 Jan 2014 19:12:17 +0000 (14:12 -0500)]
Create new IPv6 attributes for Subnets
* Introduces two new optional attributes for Subnets:
* ipv6_ra_mode
* ipv6_address_mode
Both attributes accept the following values:
* dhcpv6-stateful
* dhcpv6-stateless
* slaac
In addition to these values, additional behaviors are specified for
when only one of the attributes is set. For example, a Neutron network
that uses a physical router as a gateway, that transmits ICMPv6 Router
Advertisement packets to configure hosts on the network will create
Neutron Subnets that have ipv6_ra_mode *not* set, and ipv6_address_mode
set to 'slaac' so that Neutron will calculate EUI64 addresses for
each port assigned to the subnet, and not spawn a Dnsmasq process.
These attributes maintain backwards compatability with the enable_dhcp
Subnet attribute, by requiring a subnet with these attributes to also
have enable_dhcp set to True.
DocImpact
Implements bp ipv6-two-attributes
Change-Id: I5b2313fff5dca1c16ff939fdc4397d7f95ba3ba5
Kevin Benton [Fri, 14 Mar 2014 19:15:18 +0000 (12:15 -0700)]
BigSwitch: Sync workaround for port del deadlock
Adds a synchronization barrier to the methods
that lock the port table to avoid the mysql
update-lock/eventlet-yield deadlock that occurs
with simultaneous port deletions.
NSX: Ensure gateway devices are usable after upgrade
The gateway device database migration upon an upgrade creates
gateway devices objects from references in network gateway
objects.
While these gateway devices are perfectly usable
in network gateways, they are not directly visible to tenants
and also cannot be updated to change details such as name,
connector type or ip as well as the client certificate.
This patch fixes the DB migration in order to ensure tenant
have access to gateway devices created after an upgrade.
This patch also modifies the l2 gateway nsxlib module to
ensure request bodies are correctly created even when not all
the attributes of a gateway device are specified.
Xuhan Peng [Fri, 7 Mar 2014 01:55:28 +0000 (20:55 -0500)]
Process ICMP type for iptables firewall
In current security group code, source_port_range_min
and source_port_range_max are used to specify icmp type
and code when security group rule protocol is icmp.
However, the code _port_arg in iptables_firewall called
by _convert_sgr_to_iptables_rules skips protocol icmp
when processing the arg. This happens to both ipv4 and
ipv6 icmp firewall rules.
This fix adds --icmp-type to iptables firewall rule when
icmp type is specified.
Li Ma [Sun, 16 Mar 2014 11:32:22 +0000 (04:32 -0700)]
ML2 plugin involves in agent_scheduler migration
In agent_scheduler migration script which creates
network-dhcp-binding table and router-l3-binding table,
ML2 plugin should be included in the plugin list.
Akihiro Motoki [Fri, 7 Mar 2014 06:58:46 +0000 (15:58 +0900)]
Avoid long transaction in plugin.delete_ports()
db_plugin.delete_ports() called plugin.delete_port() under
a transaction. It leads to long transaction if plugin.delete_port
talks with external systems. This commit changes each delete_port
outside of a transaction to avoid longer transaction.
plugin.delete_ports is now called by release_dhcp_ports and
dhcp-agent ports can be deleted separately, so this changes
does not break the existing behavior.
delete_ports is renamed to delete_ports_by_device_id
to clarify the usage of this method.
NEC plugin already has this change and it is no longer needed.
_do_side_effect helper method in test_db_plugin is renamed
to more self-descriptive name.
The workflow of creating a firewall when a router without any i/f is present in that tenant
causes a KeyError on the FWAgent. The issue occurs as such routers are present in the list of
routers returned by get_routers() but are not populated in the router_info dict. Adding a check
before accessing the dict to prevent the exception. When an i/f is added to such routers -
_router_added processing in the l3agent will populate the router_info dict and the FWAgent
also picks this up in that context.
NSX: fix intermetting UT failure on vshield test_router_create
Logstash shows this happened twice during the past 24 hours.
Let's be proactive and see if we can mitigate the random failure
by raising the poll time.
Aleks Chirko [Tue, 26 Nov 2013 14:22:57 +0000 (16:22 +0200)]
Bugfix and refactoring for ovs_lib flow methods
Remove hardcoded flow parameters from
'_build_flow_expr_str' method, so we can
define any flows we want and can rely on 'ovs-ofctl'
command to verify flow arguments correctness.
When building flow string inside _build_flow_expr_str
use the following approach:
1. Build prefix and remove prefix params from flow_dict.
2. Build postfix (actions) and remove 'actions' from
flow dict.
3. Inside the loop build flow array from everything
what's left in flow_dict.
4. Append postfix (actions) to the flow array.
5. 'Join' flow array into flow string.
Change _build_flow_expr_str() to be a function
instead of an object method because 'self'
parameter wasn't used.
Remove 'add_or_mod_flow_str' method because
we have to use separate logic when bulding flow
strings for 'add_flow' and 'mod_flow' methods.
mathieu-rohon [Fri, 14 Mar 2014 09:17:55 +0000 (10:17 +0100)]
Send fdb remove message when a port is migrated
the fdb_remove rpc message is sent when the status
of the port goes to BUILD, that is when the new host
send a get_device_details which means that it owns
the migrated port. The fdb_add message will be sent
as soon as the new host send update_device_up
Code Review / openstack-build / neutron-build.git / log