NVIDIA DeepStream SDK API Reference

9.2 Release
ModelMuxRoleAttach Struct Reference

Detailed Description

One role branch (Primary or Shadow) attached for a stream: the model this role is currently routed to + every element/pad of its lane, recorded so the lane can later be swapped (model/routing) or torn down surgically.

Definition at line 733 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

Collaboration diagram for ModelMuxRoleAttach:

Data Fields

gboolean active
 
gboolean busy
 
gchar * model
 a zero-drop swap/clear probe is in flight
More...
 
ModelPool * role
 model name routed for this role
More...
 
GstElement * conv
 owning role bin
More...
 
GstElement * in_q
 OPTIONAL copy converter: tee -> nvvideoconvert -> in_q (NULL = zero-copy passthrough lane). More...
 
GstElement * out_q
 queue: [conv ->] in_q -> ModelBin (persistent on swap) More...
 
GstPad * tee_src
 queue: ModelBin -> combined mux (persistent) More...
 
GstPad * out_mux_sink
 requested src pad on the per-stream tee
More...
 
gint out_idx
 requested sink pad on combined display mux
More...
 
GstPad * tap_pad
 recycled combined-mux sink index
More...
 
gulong tap_probe_id
 debug tap probe pad (held until teardown)
More...
 
GstPad * verify_head_pad
 debug tap probe id
More...
 
gulong verify_head_probe_id
 debug copy-verify head probe pad
More...
 
GstPad * verify_queue_pad
 
gulong verify_queue_probe_id
 debug copy-verify queue probe pad
More...
 
gpointer verify_ctx
 
gboolean passthru
 debug copy-verify context
More...
 
gchar * pending_model
 model-sharding=0 overflow: this branch bypasses inference entirely – tee -> queue -> out_mux, no model bin (role/model are NULL). More...
 
gint req_gpu
 when passthru AND awaiting deferred promotion: the model this stream is MAPPED to but which is still warming. More...
 
gboolean via_default
 requested instance gpu for this role's model (MM_GPU_ANY = no constraint). More...
 

Field Documentation

◆ active

gboolean ModelMuxRoleAttach::active

◆ busy

gboolean ModelMuxRoleAttach::busy

◆ conv

GstElement * ModelMuxRoleAttach::conv

owning role bin

Definition at line 739 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ in_q

GstElement * ModelMuxRoleAttach::in_q

OPTIONAL copy converter: tee -> nvvideoconvert -> in_q (NULL = zero-copy passthrough lane).

buffer-copy-mode. Rides WITH this record on swap, like in_q/out_q.

Definition at line 742 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ model

gchar * ModelMuxRoleAttach::model

a zero-drop swap/clear probe is in flight

Definition at line 737 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ out_idx

gint ModelMuxRoleAttach::out_idx

requested sink pad on combined display mux

Definition at line 746 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ out_mux_sink

GstPad * ModelMuxRoleAttach::out_mux_sink

requested src pad on the per-stream tee

Definition at line 745 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ out_q

GstElement * ModelMuxRoleAttach::out_q

queue: [conv ->] in_q -> ModelBin (persistent on swap)

Definition at line 743 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ passthru

gboolean ModelMuxRoleAttach::passthru

debug copy-verify context

Definition at line 754 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ pending_model

gchar * ModelMuxRoleAttach::pending_model

model-sharding=0 overflow: this branch bypasses inference entirely – tee -> queue -> out_mux, no model bin (role/model are NULL).

Frames flow downstream un-inferred (nvinfer-style over-batch).

Definition at line 758 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ req_gpu

gint ModelMuxRoleAttach::req_gpu

when passthru AND awaiting deferred promotion: the model this stream is MAPPED to but which is still warming.

Shown in the routing table as "<model> (passthru)". NULL for a permanent/overflow passthrough (no intended model).

Definition at line 762 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ role

ModelPool * ModelMuxRoleAttach::role

model name routed for this role

Definition at line 738 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ tap_pad

GstPad * ModelMuxRoleAttach::tap_pad

recycled combined-mux sink index

Definition at line 747 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ tap_probe_id

gulong ModelMuxRoleAttach::tap_probe_id

debug tap probe pad (held until teardown)

Definition at line 748 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ tee_src

GstPad * ModelMuxRoleAttach::tee_src

queue: ModelBin -> combined mux (persistent)

Definition at line 744 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ verify_ctx

gpointer ModelMuxRoleAttach::verify_ctx

◆ verify_head_pad

GstPad * ModelMuxRoleAttach::verify_head_pad

debug tap probe id

Definition at line 749 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ verify_head_probe_id

gulong ModelMuxRoleAttach::verify_head_probe_id

debug copy-verify head probe pad

Definition at line 750 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ verify_queue_pad

GstPad * ModelMuxRoleAttach::verify_queue_pad

◆ verify_queue_probe_id

gulong ModelMuxRoleAttach::verify_queue_probe_id

debug copy-verify queue probe pad

Definition at line 752 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.

◆ via_default

gboolean ModelMuxRoleAttach::via_default

requested instance gpu for this role's model (MM_GPU_ANY = no constraint).

Recorded on every EXPLICIT attach / route / promote and REUSED by the internal re-attach paths (swap, recreate, promote fallback), so a stream never silently migrates off the device the operator asked for.

Definition at line 767 of file sources/gst-plugins/gst-nvmodelmux/gstnvmodelmux_priv.h.


The documentation for this struct was generated from the following file: