From my VM, if I try the metadata server repeatedly, it eventually gets a response:
albertb@<html><body><h1>503:~ $ curl http://169.254.169.254/2009-04-04/meta-data/instance-id <html><body><h1>503 Service Unavailable</h1> No server is available to handle this request. </body></html> albertb@<html><body><h1>503:~ $ curl http://169.254.169.254/2009-04-04/meta-data/instance-id <html><body><h1>504 Gateway Time-out</h1> The server didn't respond in time. </body></html> albertb@<html><body><h1>503:~ $ curl http://169.254.169.254/2009-04-04/meta-data/instance-id i-000017ccalbertb@<html><body><h1>503:~ $ Then I see this in neutron-metadata-agent.log: 2019-12-09 12:41:55.213 266833 INFO eventlet.wsgi.server [-] 10.195.72.75,<local> "GET /2009-04-04/meta-data/instance-id HTTP/1.1" status: 200 len: 146 time: 0.3922560 When it fails (503/504) nothing is logged by Neutron but I see haproxy logs: Dec 9 13:06:14 us01odc-p02-ctrl3 haproxy-metadata-proxy-569e2c42-3935-48e9-ae41-074650e57b2b[267096]: 10.195.72.75:53266 [09/Dec/2019:13:06:14.475] listener listener/metadata 0/0/-1/-1/0 503 212 - - SC-- 58/58/57/57/3 0/0 "GET /2009-04-04/meta-data/instance-id HTTP/1.1" Dec 9 13:06:59 us01odc-p02-ctrl3 haproxy-metadata-proxy-569e2c42-3935-48e9-ae41-074650e57b2b[267096]: 10.195.72.75:53268 [09/Dec/2019:13:06:27.067] listener listener/metadata 0/0/0/-1/32001 504 194 - - sH-- 90/90/89/89/0 0/0 "GET /2009-04-04/meta-data/instance-id HTTP/1.1" The load on my controllers is around 15 but they have 48 CPU so 15 should be OK. When I look at RMQ I see everything fine except for the q-plugin queue; it has 3 consumers but there are thousands of unacked messages and the number gradually increases. Could that be causing the neutron-metadata issue? From: Albert Braden <Albert.Braden@synopsys.com> Sent: Monday, December 9, 2019 12:20 PM To: openstack-discuss@lists.openstack.org Subject: RE: neutron-metadata-agent broken pipe When I try to build a VM I see this in the VM logs: 2019-12-09 20:02:21,396 - url_helper.py[WARNING]: Calling 'http://169.254.169.254/2009-04-04/meta-data/instance-id' failed [0/120s]: bad status code [503] 2019-12-09 20:03:41,084 - util.py[WARNING]: Failed fetching userdata from url http://169.254.169.254/2009-04-04/user-data 2019-12-09 12:03:53,041 - util.py[WARNING]: Failed running /var/lib/cloud/scripts/per-boot/config_instance.sh [1] 2019-12-09 12:03:53,043 - cc_scripts_per_boot.py[WARNING]: Failed to run module scripts-per-boot (per-boot in /var/lib/cloud/scripts/per-boot) This is the failing line from the script: name=`curl -s http://169.254.169.254/2009-04-04/meta-data/hostname`<http://169.254.169.254/2009-04-04/meta-data/hostname%60> When I try this from the VM I get this error: albertb@<html><body><h1>503:~ $ curl -s http://169.254.169.254/2009-04-04/meta-data/hostname <html><body><h1>503 Service Unavailable</h1> No server is available to handle this request. </body></html> When I check neutron-metadata-agent.log for the time when the VM was failing I see the "broken pipe" errors: 2019-12-09 11:56:00.075 664593 INFO eventlet.wsgi.server [-] Traceback (most recent call last): File "/usr/lib/python2.7/dist-packages/eventlet/wsgi.py", line 521, in handle_one_response write(b''.join(towrite)) File "/usr/lib/python2.7/dist-packages/eventlet/wsgi.py", line 462, in write wfile.flush() File "/usr/lib/python2.7/socket.py", line 307, in flush self._sock.sendall(view[write_offset:write_offset+buffer_size]) File "/usr/lib/python2.7/dist-packages/eventlet/greenio/base.py", line 390, in sendall tail = self.send(data, flags) File "/usr/lib/python2.7/dist-packages/eventlet/greenio/base.py", line 384, in send return self._send_loop(self.fd.send, data, flags) File "/usr/lib/python2.7/dist-packages/eventlet/greenio/base.py", line 371, in _send_loop return send_method(data, *args) error: [Errno 32] Broken pipe Why is my neutron-metadata server failing? Has anyone else seen this problem? We are running Rocky with about 200 hypervisors; it started after we added 100.