Blacklist tooz 1.57.4

tooz 1.57.4 was added to upper-constraints on 7/10 and since
then we've been seeing random thread / locking issues in the
cinder-volume service during volume create resulting in the volume
create dropping out and never reporting as completed, so tests
timeout.

The code that's involved when this blows up is shown in guru
medidation reports as going through cinder.coordination which
relies on tooz and is configured to use the etcd3 driver in tooz.

The 1.57.4 release includes two highly suspect changes for what
could be related here:

1. Make sure Lock.heartbeat() returns True/False
   I350ea10163d59a06712c22c4c5be4dfcb6885bf8

2. etcd3: replace custom lock code by more recent etcd3 lock code
   Ia5741bae83cddd90f9bb78fba9173dca8509e6d9

Given the timing of the failures with the release and all of the
related pieces involved, we shoud blacklist tooz 1.57.4 until the
tooz core team can investigate the failures and see if there is
an obvious issue, either in the tooz etcd3 locking implementation
or maybe in the cinder.coordination code. With <2 weeks to Pike
feature freeze though, we can't afford the high failure rate we're
experiencing in the CI system right now.

Change-Id: Ibde3c0fff1320a35a799ab25028d123b48e06b0b
Related-Bug: #1694509

global-requirements.txt

@@ -306,7 +306,9 @@ sympy>=0.7.6  # BSD
 taskflow>=2.7.0  # Apache-2.0
 tempest>=16.1.0  # Apache-2.0
 tempest-lib>=0.14.0  # Apache-2.0
-tooz>=1.47.0 # Apache-2.0
+# NOTE(mriedem): tooz 1.57.4 is suspected of causing locking issues
+# with the etcd3 backend, at least in cinder-volume
+tooz>=1.47.0,!=1.57.4 # Apache-2.0
 tosca-parser>=0.8.1  # Apache-2.0
 tripleo-common>=7.1.0  # Apache-2.0
 urllib3>=1.21.1  # MIT

upper-constraints.txt

@@ -317,7 +317,7 @@ os-traits===0.3.2
 SecretStorage===2.3.1
 XStatic-Rickshaw===1.5.0.0
 iso8601===0.1.11
-tooz===1.57.4
+tooz===1.57.3
 linecache2===1.0.0
 oauth2client===4.1.2
 idna===2.5