Houdini

Afanasy is represented by a special multi-functional ROP. You can connect several other ROP-s to Afanasy ROP to render. You can connect several Afanasy ROP-s to Afanasy ROP for a job with a complex dependencies.

Afanasy ROP

  • Submit
    Generate a job and send it to server.
  • Start Paused
    Send a job in off-line state.
  • Preview Approval
    Set job preview approval flag. For example, if sequential is 10, it will render every 10 frame and wait for approve.

General

../_images/houdini_afrop_general.png

Afanasy ROP General tab

  • Job Name

    Afanasy job name.

  • Output Driver

    You can not (don’t want) to connect Afanasy ROP to render ROP, you can specify it.

  • Valid Frame Range:
    • Render Any Frame
      Use frame range form downstream node. Or render current frame if no range in network defined.
    • Render Frame Range
      Render this specified frame range.
    • Render Frame Range Only (Strict)
      Render this specified frame range. Other ROP-s will wait this whole frame range rendered.
  • Single Task
    • Generate single task for whole frame range, useful for simulations.
  • Local Render

    Render on the local render client. Job host mask will be automatically set to the local host name.

  • Frames Per Task

    Number of frames in each task.

  • Sequential
    1 Render frames one by one from the first to the last
    10 Render every 10 frame at first, than render last other frames
    -1 Render frames backwards from the last to the first
    -10 Render every 10 frame at first backwards, than render last other frames backwards
    0 Render the first, the last, the middle, the middle of the middle and so on
  • Wait Time

    Set job time_wait parameter to the current day Hours and Minutes. If current time will be greater than specified, the next day will be used.

  • Render With Take

    Specify take to render.

  • Connected Nodes Are Independent

    Allow run the same frames of all connected nodes at the same time.

  • Allow Sub-Task Dependence

    Tasks can wait other tasks to be done partially. Useful for simulations. Frames render can start w/o waiting the whole simulation is finished.

  • Ignore ROP Inputs

    Do not execute input ROP-s.

Parameters

../_images/houdini_afrop_parameters.png

Afanasy ROP Parameters tab

  • Platform
    OS type the job can launch tasks on:
    • Any: any OS.
    • Native: the same as this OTL was launched on.
  • Tickets
    Use tickets:
    • Auto: Submission script will try to set tickets automatically, depending on the ROP to render.
    • Memory: If not zero, this amount of MEM tickets will be set.
    • Aux: Any other tickets string as TICKET:COUNT comma separated list.
  • Pools
    Use pools that are specified as pool:priority comma separated list.
  • Enable Extended Parameters
    To switch ON/OFF it fast.
  • Job Branch
    $HIP should be used in most cases. No matter how deep you placed hip file in file-system. It just help to find server an existing parent branch (department, project, scene).
  • Hosts Mask
    Hosts names pattern where job can run on (empty value means that job can run on host with any name).
  • Exclude
    Hosts Mask Exclude: Hosts names pattern where job can not to run on.
  • Depend Mask
    Same user jobs names pattern to wait to be done to start
  • Global
    Depend Mask Global: Same as Depend Mask, but waits for a jobs from any user.
  • Priority
    Job order in user jobs list (-1 means to use default value).
  • Maximum Running Tasks
    Maximum tasks job can run at the same time (-1 means no limit).
  • Per Host
    Maximum Running Tasks Per Host: Maximum running tasks on the same host. (-1 means no limit).
  • Capacity
    Tasks capacity value (-1 means use default value). Render must have enough free capacity to run it.
  • Render Time Min
    Minimum time task should run (seconds). Sometimes tasks finishes with a good exit status too early.
  • Max
    Task maximum running time (in hours). If task will not finish after this time, it will considered as an error and will be restarted.
  • Progress Timeout
    If a task will not produce any output for this time (in hours), it will be considered as an error.
  • Min RAM
    Minimum free memory (Gigabytes) should have render client to be able to start a task.
  • Override Service
    This will be any custom service name for a job block tasks.
  • Parser
    Override Parser: This will be any custom parser name for a job block tasks.
  • Life Time
    DONE job will be automatically deleted after this time (in hours). Useful for some auxiliary jobs.
  • Files Check
    Service (task instanced Python class) can check rendered files for existence. Submitter (script) should know file names that task should produce. Can not work on expressions/takes/overrides.
    • Skip Existing
      Render can check files for existence before run task command.

Distribute Simulation

../_images/houdini_afrop_distributed.png

Afanasy ROP Separate Render tab

  • Controls Node

    Distributed simulation control node.

  • Number Of Slices

    Distributed simulation slices number.

  • Tracker Parameters

    Distributed simulation slices tasks should communicate via tracker service.

    • Capacity
      Tracker task capacity.
    • Host Mask
      Tracker will run only hosts that names match this regular expression.
    • Service
      Tracker task service.
    • Parser
      Tracker task parser.
    • Manual Tracker
      Use manual launched tracker service at specified Address and Port

Separate Render

Separate Render allows to separate render process on IFD files generation and render it by mantra. It can give several advantages on some heavy scenes.

Separate render generates a job that can:

  • Render images locally in temporary folder and copy whole image after successful render. It can save your network traffic as render do not send small portions of an image during render process.
  • Generate IFD file locally and then render it in separate process but in the same task. It can save render memory.
  • Split one frame on tiles to render them simultaneously. So you can increase speed of one frame render. And also it can reduce memory needed to render a frame.
  • Cleanup rendered IFD files and joined tiles images.
../_images/houdini_afrop_separate.png

Afanasy ROP Separate Render tab

  • Enable Separate Render
    Turn this feature on.
  • Run ROP
    Run ROP to generate files to render. Houdini will generate IFD files for mantra.
  • Join Render Stages
    Generate IFD files and render in the same task. In this case IFD files will be generated to local temporary folder. It can save and memory usage and network traffic.
  • Read Parameters from ROP
    Read files to generate and images to render parameters from specified ROP.
  • Render Arguments
    Arguments for render command. Usually files and may be some other options.
  • Files
    Files to generate.
  • Delete ROP Files On Job Deletion
    ROP files (IFD-s) can be deleted when user will delete the job.
  • Images
    Images which render will produce. Needed for tile render, AfWatch preview/thumbnails.
  • Tile Render
    Enable rendering tiles and then combine them.
  • Divisions
    Tiles divisions.

Custom Command

Run any custom command. For example you can render IFD files using mantra command, generate a preview movie with ffmpeg.

../_images/houdini_afrop_command.png

Afanasy ROP Custom Command tab

  • Custom Command Mode
    Add custom command tasks block to a job.
  • Name
    Tasks block name. If empty the first word of the command will be used.
  • Command
    The command.
  • Prefix with $AF_CMD_PREFIX
    Add $AF_CMD_PREFIX environment variable value to the beginning of the command. This may be needed for some software (environment) setup.
  • Files
    Some files you can point to use in command.
  • Delete Files On Job Deletion
    Delete this files when user will delete job.
  • Preview
    Specify result picture here to enable tasks preview.
  • Service
    Tasks block service. If empty the first word of the command will be used.
  • Parser
    Tasks block parser.

SOHO

This can be used to explain other ROP network what to do with Afanasy node.

../_images/houdini_afrop_soho.png

Afanasy ROP SOHO tab

  • Afanasy ROP
    Specify Afanasy ROP to execute by SOHO.
  • Program
    Script that will be executed on SOHO demand. That default script will execute Submit button on a specified Afanasy ROP.

ROP Examples

Simple

Just connect afanasy ROP to your render ROP.

../_images/houdini_simple_network.png

Simple Network

../_images/houdini_simple_job.png

Simple Job

../_images/houdini_simple_tasks.png

Simple Job Tasks

The job consists of single tasks block. Each task represents a frame or several number of frames, specified in Frames Per Task parameter.

Command Render

You can send any custom command to your farm. Usually you need separate IFD files generation and run mantra as a standalone process to render.

../_images/houdini_command_network.png

Command Network

../_images/houdini_command_job.png

Command Job

../_images/houdini_command_tasks.png

Command Job Tasks

This job consists of two blocks of tasks. The first block produced by mantra_ifd node, with Disk File parameter turned on. Next block runs mantra with files parameter pointing to the generated files.

Tile Render

You can split single image to render on several hosts. Each host (task) will produce a tile - some part of an image. Tiles will be combined in a single image.

../_images/houdini_tilerender_network.png

Tile Render Network

../_images/houdini_tilerender_job.png

Tile Render Job

../_images/houdini_tilerender_tasks.png

Tile Render Job Tasks

Tile job consists of three blocks:

  • Generate
    Generate IFD files.
  • Render
    Render tiles with mantra standalone process.
  • Join
    Join tiles to assemble an image. If tiles were successfully joined they will be removed. At the end of this stage, IFD will be removed, if it was asked.

Houdini native itilestitch tool is used to join tiles.

Sub Task Dependence

This option is designed to start to render simulation without waiting the whole simulation is finished.

../_images/houdini_subtask_network.png

Sub-Task Dependence Network

The first block of a job is a simulation. It consists of a single task (Frames Per Task parameter is set to the whole frame range). The second block set to wait the first one with sub-task dependence. So it begins to render as first frames of a simulation completed, while the simulation task is still running.

../_images/houdini_subtask_job.png

Sub-Task Dependence Job

We also can notice here, that the render block got HYTHON and MANTRA tickets, while the simulation block got only HYTHON ticket

../_images/houdini_subtask_tasks.png

Sub-Task Dependence Job Tasks

Complex

You can construct a complex Afanasy ROP network to construct a complex job.

../_images/houdini_complex_network.png

Complex Network

../_images/houdini_complex_job.png

Complex Job

This job consists of a simulation with sub-task dependence. Two caches waiting the simulation, but can run independently from each other. Mantra tile render which produces three blocks which wait all the cache. Two blocks for preview which can run independently but wait tile render tasks. One to convert EXR files to JPEG-s and one to generate a preview movie form EXR-s.

Distributed Simulations

Houdini can calculate the same simulation on several machines.

How It Works

Simulation can be split on slices, so each machine calculates own slice. But different slices simulations should exchange information to pass data from slice to slice. Houdini has a special Python script simtracker.py for it. It needs to launch a server that simulations will connect to. So each slice simulation should know tracker address and port. Also tracker has a simple web interface to see logs.

What We Should Do

  • Prepare distributed simulation, setup slices.
  • Launch tracker server and get its address and port.
  • Open several Houdini applications with simulation scene (on different machines or not).
  • Specify tracker and port.
  • Start each Houdini instance to simulate own slice.
  • Stop tracker.

So, you can distribute Houdini simulation without any render farm manager.

Step-By-Step

  1. Create a sphere.

  2. Create simulation via Wispy Smoke shelf tool.

  3. Apply Distribute Container shelf tool.

  4. You will be moved to /out/ network.

  5. Create Afanasy ROP node.

  6. Set Output Driver to /obj/distribute_pyro/save_slices and in the Distributed Simulation tab set Controls Node to /obj/AutoDopNetwork/DISRIBUTE_pyro_CONTROLS. You can copy this values from HQueue Simulation ROP that was automatically created.

    ../_images/houdini_distribpyro_afgeneral.png

    Genetal Tab

    ../_images/houdini_distribpyro_afdistrib.png

    Distributed Simulation Tab

  7. Uncheck Render Temporary HIP File option on Afanasy ROP. By default, Afanasy renders a temporary scene to allow user to continue working with original file. But in this case $HIPNAME variable will change, and it widely used in shelf tools and examples.

  8. Go to /obj/AutoDopNetwork/.

  9. Remove resize_container node.

  10. Disconnect distribute_pyro node from merge node (do not merge it with source). And connect it to the solver Velocity Update input.
    ../_images/houdini_distribpyro_dop_orig.png

    Original network

    ../_images/houdini_distribpyro_dop_adjust.png

    Adjusted network

  11. Set slices divisions 1 x 2 x 1.

  12. Now you can submit simulation by Afanasy ROP in /out/ network.

Afanasy Job

Afanasy will create a job that consists of four blocks each contains just one task. First block task to start tracker. A block (task) for each slice that waits tracker start. And the last block task to stop the tracker.

../_images/houdini_distribpyro_job_running.png

Distributed Simulation Job Running

  1. tracker

    The first task block has a special service htracker. This service just adds job ID to the task command. Job ID is needed to manipulate job using JSON protocol. The command calls a special CGRU Python script plugins/houdini/htracker.py.

    htracker --start --envblocks "save_slices.*|tracker-stop" --depblocks "save_slices.*"
    • It starts Houdini simtracker in a separate thread and gets its address and port.
    • Set other job blocks environment variables TRACKER_ADDRESS and TRACKER_PORT to blocks specified by –envblocks argument.
    • Set slices job blocks depend masks to an empty string to blocks specified by –depblocks argument, So that blocks will wait nothing and can to start.
    • Waits simtracker for completion.

  2. save_slices-s0

    The first slice simulation. Slices are simulated by CGRU multi-functional Hython script cgru/plugins/houdini/hrendef_af.py that Afanasy uses for almost everything.

    hrender_af -s 1001 -e 1133 --by 1 -t "_current_" --ds_node "/obj/AutoDopNetwork/DISTRIBUTE_pyro_CONTROLS" --ds_address "localhost" --ds_port 8000 --ds_slice 0 "/opt/cgru/examples/houdini/distrib_pyro.hip" "/obj/distribute_pyro/save_slices"

    Control node, tracker address and tracker port, that was specified in Afanasy ROP and passed by command line argument, will be overridden by environment variables.

    Script will open HIP file, set control node tracker address and port parameters. Set SLICE variable to the specified slice number.

    Run simulation ROP.

  3. save_slices-s1

    The second slice simulation. It is the same as the first, but with one key difference. Slice will be equal to 1.

    hrender_af -s 1001 -e 1133 --by 1 -t "_current_" --ds_node "/obj/AutoDopNetwork/DISTRIBUTE_pyro_CONTROLS" --ds_address "localhost" --ds_port 8000 --ds_slice 1 "/opt/cgru/examples/houdini/distrib_pyro.hip" "/obj/distribute_pyro/save_slices"

  4. tracker-stop

    Stop tracker. It will be performed by the same script that starts tracker.

    htracker --stop

    It just sends quit string to tracker_address:tracker_port socket.

../_images/houdini_distribpyro_job_done.png

Distributed Job Done

../_images/houdini_distribpyro_job_stopping.png

Distributed Job Stopping

../_images/houdini_distribpyro_tasks.png

Distributed Job Tasks

Afanasy TOP Scheduler

This node executes work items on farm using Afanasy render manager. It can schedule work items from TOP UI, and as standalone job. Using a standalone job you can close Houdini session and watch progress via Afanasy GUI.

../_images/houdini_pdg_cooking1.png
../_images/houdini_pdg_cooking2.png

Scheduling from Houdini TOP UI

../_images/houdini_pdg_job1.png
../_images/houdini_pdg_job2.png
../_images/houdini_pdg_job3.png

Scheduling using a standalone job

Here is the scheduler nodes parameters description. Almost anywhere in Afanasy -1 means that the value is not set and defaults will be used.

Scheduling Parameters

../_images/houdini_pdg_scheduler.png

Afanasy TOP Shecduler tab

  • Job Name
    The name of the job where work items tasks will be appended to.
  • Job Branch
    The branch of the job. The same value will be used if you submit graph as job.
  • PDG Directory
    Specifies the directory where the cook generates intermediate files. The intermediate files are placed in a subdirectory named pdgtemp.
  • Path Mapping
    • Global
      If the PDG Path Map exists, then it is applied to file paths.
    • None
      Delocalizes paths using the PDG_DIR token.
  • Path Map Zone
    When on, specifies a custom mapping zone to apply to all jobs executed by this scheduler. Otherwise, the local platform is LINUX, MAC or WIN.

Submit Graph As Job

  • Submit Graph As Job
    Cooks the entire TOP network as a standalone job. Displays the status URI for the submitted job. The submitting Houdini session is detached from the cooking of the TOP network.
  • Start Paused
    Start graph cooking job paused.
  • Priority
    Graph cooking job priority value.
  • Capacity
    Cooking task capacity.
  • Hosts Mask
    Hosts names regular expression, where graph job can run.
  • Exclude
    Hosts names regular expression, where graph job can not run.
  • Depend Mask
    Current user jobs names expression, that job will wait to start for.
  • Global
    Any user jobs names expression, that job will wait to start for.
  • Service
    Cooking job task block service name.
  • Ticket
    Cooking job task block will need and take one ticket with this name. See tickets documentation for details.

Tasks Parameters

You can override this parameters on each TOP node, except Job Priority which will be set to an entire job.

../_images/houdini_pdg_parameters.png

Afanasy TOP Tasks Parameters tab

  • Job Priority
    Priority value of a job were working items tasks will be executed.
  • Capacity
    Work items tasks block capacity.
  • Hosts Mask
    Hosts names regular expression, where tasks can run.
  • Exclude
    Hosts names regular expression, where tasks can not run.
  • Max Running Tasks
    Running tasks count at the same time limit.
  • Per Host
    Running tasks count at the same time on the same host limit.
  • Render Time Min (Sec)
    Minimum task running time limit. If task will finish for seconds below this value, task finish will be considered as with an error.
  • Max (Hours)
    Maximum task running time limit. If task will run for hours above this value, it will be forced to stop with an error.
  • Min RAM (GB)
    Host should have this count of Gigabytes of a free RAM to be able to run tasks.
  • Service
    Tasks block service. If empty it will try to detect automatically. If node fetches ifd ROP, service will be hbatch_mantra, ffmpegencodevideo will be ffmpeg.
  • Tickets
    asks block tickets. A comma separated list of key:count. Example: MEM:64,GPU:1. See tickets documentation for details.
  • Auto
    Automatically add common tickets. Almost all tasks launch hython, so HYTHON ticket will be added. If node fetches ifd ROP, MANTRA ticket will be added.
  • Environment
    Adds custom key-value environment variables to tasks block.

To override task parameter on TOP node add it via Edit Parameter Interface window:

../_images/houdini_pdg_edit_parameter_interface.png

Edit Parameter Interface window

Adjustment Parameters

../_images/houdini_pdg_adjustment.png

Afanasy TOP Adjustment tab

  • Report Item Fail On Error

    If task gets error state, scheduler will report PDG that work item failed. You can turn it off and try to solve errors via Afanasy only. Read output for an error cause, try to fix it, restart task. And PDG will know nothing about it.

  • Block On Failed Work Items

    When this option is enabled the scheduler will block the cook from completing if there are any failed work items in that scheduler. This makes it possible to manually retry those work items, by preventing the PDG graph cook from ending before failed items can be retried. A cook that is blocked on failed work items can still be canceled using the ESC key, the cancel button in the TOP task bar, or the cancel API method.

  • Validate Outputs When Recooking

    When enabled, PDG will check the output files of work items when the graph recooks, to see if the files still exist on disk. Work items that are missing output files will be dirtied and cook again.

  • Check Expected Outputs On Disk

    When enabled, PDG will look on disk for any expected work items outputs that were not explicitly reported when the work item cooked. Expected outputs for a work item are checked immediately after the scheduler marks the work item as cooked. Output files that were reported by the work item normally while cooking will not be checked.

  • Use IP Address

    Use IP address instead of host name as work item result server address. Some times render farm can’t solve workstations by name. Also it can save DNS load.

    Work item task can send progress to PDG itself. It is used by batch work item to notify that a specific frame (item in the batch) is done. This way PDG can start to render images when just first frames of a simulation rendered, and there is no need to wait the entire simulation finish. So work item should know address and port to send progress to.

  • Tick Period

    Sets the minimum time (in seconds) between calls to the onTick callback. This callback is called periodically when the graph is cooking. The callback is generally used to check the state of running work items.

    Afanasy server heart beat is 1 second, so there is no sense to set this parameter less than a second.

  • Max Items Per Tick

    Sets the maximum number of ready item onSchedule callbacks between ticks.

    For example by default the tick period is 1s and the max items per tick is 30. This means that scheduler can send a maximum of 30 work items per second to farm. Adjusting these values can be useful to control the load on the farm scheduler.

Setup

CGRU setup should be sourced before. To do this you can source setup.sh script in CGRU root folder. Afanasy Houdini operator library and Python module are located in:

cgru/plugins/houdini

You should add this folder HOUDINI_PATH and PYTHONPATH environment variables.

Houdini setup example (bash):

# Setup CGRU
cd /opt/cgru
source ./setup.sh

# Setup CGRU houdini plugins location:
export HOUDINI_CGRU_PATH="${CGRU_LOCATION}/plugins/houdini"

# Append HOUDINI_PATH with CGRU plugins:
export HOUDINI_PATH="${HOUDINI_CGRU_PATH}:&"

# Append Python path with afanasy submission script:
export PYTHONPATH="${HOUDINI_CGRU_PATH}:${PYTHONPATH}"

If you avoid sourcing cgru/setup.sh see Manual Environment Setup.