Multiple-Reader/Single-Writer locks, or "Busy" flag for workspace access

[Janik Zikovsky]

Refs #4516: It is possible to have destructive access to a workspace (via the PlusMD algorithm, for example), while the GUI thread is busy accessing it to draw it.

We can prevent segfaults by either: implementing a "busy" flag on workspaces, or by using Multiple-Reader/Single-Writer locks.

​http://stackoverflow.com/questions/989795/example-for-boost-shared-mutex-multiple-reads-one-write

Investigate.

Edit: Here is a sample script that could cause segfaults if using GUI views on the MDW workspace while the loop runs:

import numpy as np

CreateMDWorkspace(Dimensions='3',Extents='0,10,0,10,0,10',Names='x,y,z',Units='m,m,m',MaxRecursionDepth='20',OutputWorkspace='mdw', SplitInto='2', SplitThreshold='150')
FakeMDEventData(InputWorkspace='mdw',UniformParams='1e+05')
plotSlice('mdw')

num = 100
xs = np.random.rand(num)*10.0
ys = np.random.rand(num)*10.0
zs = np.random.rand(num)*10.0
for (x,y,z) in zip(xs, ys, zs):
	CreateMDWorkspace(Dimensions='3',Extents='0,10,0,10,0,10',Names='x,y,z',Units='m,m,m',MaxRecursionDepth='20',OutputWorkspace='extra')
	FakeMDEventData(InputWorkspace='extra',PeakParams='1e4, %g,%g,%g, 0.5'% (x,y,z))
	PlusMD('mdw', 'extra', OutputWorkspace='mdw')
	PlusMD('mdw', 'extra', OutputWorkspace='mdw')
	PlusMD('mdw', 'extra', OutputWorkspace='mdw')
	PlusMD('mdw', 'extra', OutputWorkspace='mdw')

[Janik Zikovsky]

Refs #4554 trying a boost mutex class, mostly to see if it compiles

Changeset: 5dd3ae31ee4b795c4b87b72e6d081148a8b669ad

[Janik Zikovsky]

Refs #4554 boost-thread not on windows

Changeset: 88aca9f20aae7e815ec0607cf38c60dd4189d336

[Janik Zikovsky]

Refs #4554 boost-thread not on windows

Changeset: fec69d95b298326197c3c41b01bbeaca34cca779

[Janik Zikovsky]

Refs #4554 commented-out test that works for me

Changeset: a44b838fa7322ce14675c456623aa891e0620f35

[Janik Zikovsky]

Refs #4554 using Poco::RWLock in a test

Changeset: 12d815690785cb94b9ba3cb94e6a74bb3e0aa20d

[Janik Zikovsky]

Refs #4554 Public ReadLock and private lock access

Also Refs #996: starting to clean up Algorithm.execute()

Changeset: be86fc7198eee110c3d262a25be9213da1b5de06

[Janik Zikovsky]

Refs #4554 make sure all destructors are virtual and ctors are called

To fix segfaults on destruction. Multiple virtual inheritance is hard ;)

Changeset: 272de2602728e2b4d4b9855305430e0911503757

[Janik Zikovsky]

Refs #4554 notes for later

Changeset: 1a6622b9c614649f144deca3e9d25b0292c880b9

[Janik Zikovsky]

Refs #4554 read/write locks on workspaces in Algorithm.

Also refs #996 cleanup

Changeset: ff60eaed87ed19b698f18594b4365b4b9c0568fc

[Janik Zikovsky]

Refs #4554 make sure to have a unique lock object

when cloning a workspace. So explicitely declared the copy constructors of every class up to DataItem.

Changeset: d6ae8f24b99c22ec2c81cbc597267a1d8617eec1

[Janik Zikovsky]

Refs #4554 left some debugging statements in

Changeset: c515accbeb966b87ef4eb6ebb9d9b2c2a21e7239

[Janik Zikovsky]

Refs #4554 added locks to Paraview renderers

To prevent segfaults from algorithms running in the background

Changeset: bb5b66e79c0ab9846386a07909afcb42c7488ca4

[Janik Zikovsky]

Note to testers: it should be possible to view a MD workspace in VatesSimpleGUI and continually update the settings (so that it fetches the data from the workspace) WHILE the above script runs in the background. The plotting blocks WS modification and vice/versa.

[Janik Zikovsky]

Refs #4554 added locks to Paraview renderers

To prevent segfaults from algorithms running in the background

Changeset: bb5b66e79c0ab9846386a07909afcb42c7488ca4

[Janik Zikovsky]

Refs #4554 debugging output for locking

Changeset: dc5a5b22ded47a49b80cb6ffd7d990d697b74eb4

[Mathieu Doucet]

Re #4554 fix warning

Changeset: 36bc9c4cd1440fde6d1afe7ceadb8141da180d2d

[Vickie Lynch]

Locks work and there is no SegFault in script

[Stuart Campbell]

This ticket has been transferred to github issue ​5401