yahoo-eng-team team mailing list archive

[siyingchun <1652919@xxxxxxxxxxxxxxxxxx>]

Public bug reported:

In our Newton test environment, we sometimes found that all instances in
the same host can correctly acquire the internal ip addresses, but they
can't be accessed by others belonging to the same tenant, meanwhile they
can't access others if you use VNC portal to login its terminal.

In order to work correctly, we could restart the ovs-agent, it can work
some time, maybe a week, maybe more, but suddenly it will come again.
Finally we find it's because br-int loses a very important flow which is
"action=NORMAL" in table 0.

When checking the code of ovs-agent, we find only table 23 is checked
and don't care about table 0, so it will terribly happen, because of
unexpected reasons.


* Precondition: 
You'd better have a large scale environment, or a small test one may be ok.
Admin account is available 

* Step-by-step:
1. Go to horizon and login with admin account 
2. Go to compute > instance > filter all instances in the same host 
3. Go to vnc for each of them, ifconfig 
4. Check internal ip, you can find them
5. Do ping others or internet

* Expect result: 
The internal ip can be allocated correctly and can access others successfully.

* Actual result: 
The internal ip can be allocated correctly, but can't access others or be accessed.

** Affects: neutron
     Importance: Undecided
     Assignee: siyingchun (wintersi)
         Status: New

** Changed in: neutron
     Assignee: (unassigned) => siyingchun (wintersi)

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1652919

Title:
  The internal ip can be allocated correctly, but the instance can't
  access others or be accessed

Status in neutron:
  New

Bug description:
  In our Newton test environment, we sometimes found that all instances
  in the same host can correctly acquire the internal ip addresses, but
  they can't be accessed by others belonging to the same tenant,
  meanwhile they can't access others if you use VNC portal to login its
  terminal.

  In order to work correctly, we could restart the ovs-agent, it can
  work some time, maybe a week, maybe more, but suddenly it will come
  again. Finally we find it's because br-int loses a very important flow
  which is "action=NORMAL" in table 0.

  When checking the code of ovs-agent, we find only table 23 is checked
  and don't care about table 0, so it will terribly happen, because of
  unexpected reasons.

  
  * Precondition: 
  You'd better have a large scale environment, or a small test one may be ok.
  Admin account is available 

  * Step-by-step:
  1. Go to horizon and login with admin account 
  2. Go to compute > instance > filter all instances in the same host 
  3. Go to vnc for each of them, ifconfig 
  4. Check internal ip, you can find them
  5. Do ping others or internet

  * Expect result: 
  The internal ip can be allocated correctly and can access others successfully.

  * Actual result: 
  The internal ip can be allocated correctly, but can't access others or be accessed.

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1652919/+subscriptions