[sirikata/sirikata] 829ce4: Disable debug prints in MeshAggregateManager.
0 views
Skip to first unread message
GitHub
May 13, 2013, 4:52:07 PM5/13/13
to sirikata...@googlegroups.com
Branch: refs/heads/master
Home: https://github.com/sirikata/sirikata
Commit: 829ce43ed48867a7f8bd4cc772065eef0299b0fe
https://github.com/sirikata/sirikata/commit/829ce43ed48867a7f8bd4cc772065eef0299b0fe
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-04-12 (Fri, 12 Apr 2013)
Changed paths:
M libspace/plugins/mesh/MeshAggregateManager.cpp
Log Message:
-----------
Disable debug prints in MeshAggregateManager.
Commit: 10c12b9483d5c855788b558ad64b8833aefa3507
https://github.com/sirikata/sirikata/commit/10c12b9483d5c855788b558ad64b8833aefa3507
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-04-13 (Sat, 13 Apr 2013)
Changed paths:
M libspace/include/sirikata/space/LocationService.hpp
M libspace/plugins/prox/LibproxManualProximity.cpp
M libspace/plugins/prox/LibproxProximity.cpp
M libspace/plugins/prox/LibproxProximityBase.cpp
M libspace/plugins/prox/LibproxProximityBase.hpp
M libspace/plugins/standard/AlwaysLocationUpdatePolicy.cpp
M libspace/plugins/standard/AlwaysLocationUpdatePolicy.hpp
M libspace/src/LocationService.cpp
Log Message:
-----------
Move all subscription posting into LocationUpdatePolicy.
Instead of having the Proximity implementations deal with posting,
have LocationUpdatePolicy manage the process. This way, it can decide
to just directly post, handle requests immediately, etc. -- it can
decide how to handle requests most efficiently.
Commit: 5b65648c292a536b10209fadd797bc44f2ef4993
https://github.com/sirikata/sirikata/commit/5b65648c292a536b10209fadd797bc44f2ef4993
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-04-13 (Sat, 13 Apr 2013)
Changed paths:
M libspace/plugins/standard/AlwaysLocationUpdatePolicy.cpp
M libspace/plugins/standard/AlwaysLocationUpdatePolicy.hpp
Log Message:
-----------
Handle (un)subscription requests in AlwaysLocationUpdatePolicy more efficiently.
Instead of passing all requests with post(), using the main strand as
an implicit queue, explicitly store requests, handle them quickly if
they reverse a still-waiting earlier request, only post requests for
processing when one isn't already running, and handle them in batches
so there isn't so much overhead to do minor modifications to the
subscription data structures. The tradeoff is that we now do more
checks up front in the query processing thread. However, this should
improve efficiency in one particular case where we block up the main
strand by posting so may tasks that take a long time to process when a
query is first evaluated and all parts of the tree up to the cut need
to be added and immediately removed.
Commit: 4a0f82312c5e8dd52d7e9eb32600999b3417c193
https://github.com/sirikata/sirikata/commit/4a0f82312c5e8dd52d7e9eb32600999b3417c193
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-04-23 (Tue, 23 Apr 2013)
Changed paths:
M libcore/include/sirikata/core/network/SSTImpl.hpp
M liboh/plugins/manual_query/ManualObjectQueryProcessor.cpp
M liboh/plugins/simple_query/SimpleObjectQueryProcessor.cpp
Log Message:
-----------
Longer initial RTO and logging.
Make the initial RTO for connections in SST longer by just multiplying
the value. This makes sure the initial value isn't too low an
approximation if the first round trip happens to be very quick due to
low load (both network and CPU). If it remains as low as the initial
value, it will quickly converge anyway.
Also add a couple of logging statements at insane level for experiments.
Commit: 5262b874201b49ea86dc1adb9be7e5e2e5226328
https://github.com/sirikata/sirikata/commit/5262b874201b49ea86dc1adb9be7e5e2e5226328
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-04-25 (Thu, 25 Apr 2013)
Changed paths:
M liboh/plugins/js/JSObjectStructs/JSTimerStruct.cpp
Log Message:
-----------
Fix clearing of GC'd timers.
"Disable" timers when clearing them when the have no references left
because the function itself uses the liveness lock, which then blocks
proper destruction, getting it stuck in an infinite loop.
Commit: 60132a14f74e3a50534429a2402bd6777487ab6f
https://github.com/sirikata/sirikata/commit/60132a14f74e3a50534429a2402bd6777487ab6f
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-05-03 (Fri, 03 May 2013)
Changed paths:
M liboh/plugins/manual_query/ServerQueryHandler.cpp
M liboh/src/HostedObject.cpp
M liboh/src/ObjectHost.cpp
M libspace/plugins/prox/CBRLocationServiceCache.cpp
M libspace/plugins/prox/CBRLocationServiceCache.hpp
M libspace/plugins/prox/LibproxManualProximity.cpp
M libspace/plugins/prox/LibproxProximity.cpp
Log Message:
-----------
Fix bad interaction between switching static/dynamic trees and migration.
When we detect that something should move between the static and
dynamic trees, we save a record and process the move on the next
tick. However, it is possible the object will migrate in the meantime,
leaving a record of the switch but no object to operate on. Removing
the record when the object migrates would be ideal, but is a difficult
change to make because of synchronization. Instead, expose a function
in the location service cache to check whether an object still exists
or not and only process the move if the object is still available.
This also adjusts CBRLocationServiceCache to track existence as a
reference count to handle migrations. Because events can come in from
different servers in the "wrong" order (i.e. get replica from server
B where object O has moved to, then get the replica removal from
server A where it moved from), we need to use counts to track whether
we're aware of *any* valid copies of an object since they all share
one record in the location service cache.
Commit: 3d8606194adad53a76447e90070ef42e24d01b75
https://github.com/sirikata/sirikata/commit/3d8606194adad53a76447e90070ef42e24d01b75
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-05-07 (Tue, 07 May 2013)
Changed paths:
M externals/prox
M libcore/include/sirikata/core/util/Vector4.hpp
M libmesh/plugins/collada/MeshdataToCollada.cpp
M libmesh/plugins/nvtt/TextureAtlasFilter.cpp
M libmesh/src/MeshSimplifier.cpp
M libspace/plugins/mesh/MeshAggregateManager.cpp
M libspace/plugins/mesh/MeshAggregateManager.hpp
Log Message:
-----------
Merge remote branch 'origin/master'
Conflicts:
libspace/plugins/mesh/MeshAggregateManager.cpp
Commit: e946ef8145eda5d3ce9734097d4a000f08dfa7c6
https://github.com/sirikata/sirikata/commit/e946ef8145eda5d3ce9734097d4a000f08dfa7c6
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-05-13 (Mon, 13 May 2013)
Changed paths:
M libspace/plugins/mesh/MeshAggregateManager.cpp
Log Message:
-----------
Fix missing start-comment in MeshAggregateManager, probably lost during the merge.
Compare: https://github.com/sirikata/sirikata/compare/20a5119494a4...e946ef8145ed
Home: https://github.com/sirikata/sirikata
Commit: 829ce43ed48867a7f8bd4cc772065eef0299b0fe
https://github.com/sirikata/sirikata/commit/829ce43ed48867a7f8bd4cc772065eef0299b0fe
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-04-12 (Fri, 12 Apr 2013)
Changed paths:
M libspace/plugins/mesh/MeshAggregateManager.cpp
Log Message:
-----------
Disable debug prints in MeshAggregateManager.
Commit: 10c12b9483d5c855788b558ad64b8833aefa3507
https://github.com/sirikata/sirikata/commit/10c12b9483d5c855788b558ad64b8833aefa3507
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-04-13 (Sat, 13 Apr 2013)
Changed paths:
M libspace/include/sirikata/space/LocationService.hpp
M libspace/plugins/prox/LibproxManualProximity.cpp
M libspace/plugins/prox/LibproxProximity.cpp
M libspace/plugins/prox/LibproxProximityBase.cpp
M libspace/plugins/prox/LibproxProximityBase.hpp
M libspace/plugins/standard/AlwaysLocationUpdatePolicy.cpp
M libspace/plugins/standard/AlwaysLocationUpdatePolicy.hpp
M libspace/src/LocationService.cpp
Log Message:
-----------
Move all subscription posting into LocationUpdatePolicy.
Instead of having the Proximity implementations deal with posting,
have LocationUpdatePolicy manage the process. This way, it can decide
to just directly post, handle requests immediately, etc. -- it can
decide how to handle requests most efficiently.
Commit: 5b65648c292a536b10209fadd797bc44f2ef4993
https://github.com/sirikata/sirikata/commit/5b65648c292a536b10209fadd797bc44f2ef4993
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-04-13 (Sat, 13 Apr 2013)
Changed paths:
M libspace/plugins/standard/AlwaysLocationUpdatePolicy.cpp
M libspace/plugins/standard/AlwaysLocationUpdatePolicy.hpp
Log Message:
-----------
Handle (un)subscription requests in AlwaysLocationUpdatePolicy more efficiently.
Instead of passing all requests with post(), using the main strand as
an implicit queue, explicitly store requests, handle them quickly if
they reverse a still-waiting earlier request, only post requests for
processing when one isn't already running, and handle them in batches
so there isn't so much overhead to do minor modifications to the
subscription data structures. The tradeoff is that we now do more
checks up front in the query processing thread. However, this should
improve efficiency in one particular case where we block up the main
strand by posting so may tasks that take a long time to process when a
query is first evaluated and all parts of the tree up to the cut need
to be added and immediately removed.
Commit: 4a0f82312c5e8dd52d7e9eb32600999b3417c193
https://github.com/sirikata/sirikata/commit/4a0f82312c5e8dd52d7e9eb32600999b3417c193
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-04-23 (Tue, 23 Apr 2013)
Changed paths:
M libcore/include/sirikata/core/network/SSTImpl.hpp
M liboh/plugins/manual_query/ManualObjectQueryProcessor.cpp
M liboh/plugins/simple_query/SimpleObjectQueryProcessor.cpp
Log Message:
-----------
Longer initial RTO and logging.
Make the initial RTO for connections in SST longer by just multiplying
the value. This makes sure the initial value isn't too low an
approximation if the first round trip happens to be very quick due to
low load (both network and CPU). If it remains as low as the initial
value, it will quickly converge anyway.
Also add a couple of logging statements at insane level for experiments.
Commit: 5262b874201b49ea86dc1adb9be7e5e2e5226328
https://github.com/sirikata/sirikata/commit/5262b874201b49ea86dc1adb9be7e5e2e5226328
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-04-25 (Thu, 25 Apr 2013)
Changed paths:
M liboh/plugins/js/JSObjectStructs/JSTimerStruct.cpp
Log Message:
-----------
Fix clearing of GC'd timers.
"Disable" timers when clearing them when the have no references left
because the function itself uses the liveness lock, which then blocks
proper destruction, getting it stuck in an infinite loop.
Commit: 60132a14f74e3a50534429a2402bd6777487ab6f
https://github.com/sirikata/sirikata/commit/60132a14f74e3a50534429a2402bd6777487ab6f
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-05-03 (Fri, 03 May 2013)
Changed paths:
M liboh/plugins/manual_query/ServerQueryHandler.cpp
M liboh/src/HostedObject.cpp
M liboh/src/ObjectHost.cpp
M libspace/plugins/prox/CBRLocationServiceCache.cpp
M libspace/plugins/prox/CBRLocationServiceCache.hpp
M libspace/plugins/prox/LibproxManualProximity.cpp
M libspace/plugins/prox/LibproxProximity.cpp
Log Message:
-----------
Fix bad interaction between switching static/dynamic trees and migration.
When we detect that something should move between the static and
dynamic trees, we save a record and process the move on the next
tick. However, it is possible the object will migrate in the meantime,
leaving a record of the switch but no object to operate on. Removing
the record when the object migrates would be ideal, but is a difficult
change to make because of synchronization. Instead, expose a function
in the location service cache to check whether an object still exists
or not and only process the move if the object is still available.
This also adjusts CBRLocationServiceCache to track existence as a
reference count to handle migrations. Because events can come in from
different servers in the "wrong" order (i.e. get replica from server
B where object O has moved to, then get the replica removal from
server A where it moved from), we need to use counts to track whether
we're aware of *any* valid copies of an object since they all share
one record in the location service cache.
Commit: 3d8606194adad53a76447e90070ef42e24d01b75
https://github.com/sirikata/sirikata/commit/3d8606194adad53a76447e90070ef42e24d01b75
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-05-07 (Tue, 07 May 2013)
Changed paths:
M externals/prox
M libcore/include/sirikata/core/util/Vector4.hpp
M libmesh/plugins/collada/MeshdataToCollada.cpp
M libmesh/plugins/nvtt/TextureAtlasFilter.cpp
M libmesh/src/MeshSimplifier.cpp
M libspace/plugins/mesh/MeshAggregateManager.cpp
M libspace/plugins/mesh/MeshAggregateManager.hpp
Log Message:
-----------
Merge remote branch 'origin/master'
Conflicts:
libspace/plugins/mesh/MeshAggregateManager.cpp
Commit: e946ef8145eda5d3ce9734097d4a000f08dfa7c6
https://github.com/sirikata/sirikata/commit/e946ef8145eda5d3ce9734097d4a000f08dfa7c6
Author: Ewen Cheslack-Postava <ewe...@cs.stanford.edu>
Date: 2013-05-13 (Mon, 13 May 2013)
Changed paths:
M libspace/plugins/mesh/MeshAggregateManager.cpp
Log Message:
-----------
Fix missing start-comment in MeshAggregateManager, probably lost during the merge.
Compare: https://github.com/sirikata/sirikata/compare/20a5119494a4...e946ef8145ed
Reply all
Reply to author
Forward
0 new messages