B !`O@sdZddlmZddlmZddlmZddlmZddlmZddlm Z ddl m Z dd l m Z dd lmZd d Zd dZGddde jZdS)z#The TensorBoard Debugger V2 plugin.)absolute_import)division)print_function)wrappers)errors) plugin_util) base_plugin)debug_data_provider) http_utilcCstj|d|idddS)Nerrorzapplication/jsoni)code)r Respond)request error_messagera/home/dcms/DCMS/lib/python3.7/site-packages/tensorboard/plugins/debugger_v2/debugger_v2_plugin.py_error_responsesrcCs t|dS)Nzrun parameter is not provided)r)rrrr_missing_run_error_response$srcseZdZdZejZfddZddZddZ dd Z e j j d d Ze j j d d Ze j j ddZe j j ddZe j j ddZe j j ddZe j j ddZe j j ddZe j j ddZe j j ddZe j j ddZZS) DebuggerV2Pluginz#Debugger V2 Plugin for TensorBoard.cs*tt|||j|_t|j|_dS)zInstantiates Debugger V2 Plugin via TensorBoard core. Args: context: A base_plugin.TBContext instance. N)superr__init__ZlogdirZ_logdirr ZLocalDebuggerV2DataProvider_data_provider)selfcontext) __class__rrr-szDebuggerV2Plugin.__init__c Cs2|j|j|j|j|j|j|j|j|j|j |j d S)N) z/runsz/alertsz/execution/digestsz/execution/dataz/graph_execution/digestsz/graph_execution/dataz/graphs/graph_infoz/graphs/op_infoz/source_files/listz/source_files/filez/stack_frames/stack_frames) serve_runs serve_alertsserve_execution_digestsserve_execution_dataserve_graph_execution_digestsserve_graph_execution_dataserve_graph_infoserve_graph_op_infoserve_source_files_listserve_source_fileserve_stack_frames)rrrrget_plugin_apps;sz DebuggerV2Plugin.get_plugin_appscCst|jjddS)ahCheck whether the Debugger V2 Plugin is always active. When no data in the tfdbg v2 format is available, a custom information screen is displayed to instruct the user on how to generate such data to be able to use the plugin. Returns: `True` if and only if data in tfdbg v2's DebugEvent format is available. ) experiment_id)boolr list_runs)rrrr is_activeKs zDebuggerV2Plugin.is_activecCstjddddS)NTz Debugger V2F)Zis_ng_componentZtab_nameZdisable_reload)rZFrontendMetadata)rrrrfrontend_metadataWsz"DebuggerV2Plugin.frontend_metadatacCsLt|j}|jj|d}t}x|D]}d|ji||j<q&Wt ||dS)N)r( start_timezapplication/json) rr(environrr*dictr-Zrun_idr r )rr experimentrunsZ run_listingrunrrrr\s   zDebuggerV2Plugin.serve_runsc Cst|j}|jd}|dkr(t|St|jdd}t|jdd}|jdd}tj||||d}|j j ||j |d}t t |j} y&t||j j||| d jd d Stjk r} zt|t| Sd} ~ XYnXdS) Nr2begin0endz-1 alert_type)r6)r( plugin_namerun_tag_filterr)blob_keyzapplication/json)rr(r.argsgetrintr Zalerts_run_tag_filterrread_blob_sequencesr7nextitertagsr r read_blobr9rInvalidArgumentErrorrstr) rrr0r2r3r5r6r8blob_sequencestagerrrres,  zDebuggerV2Plugin.serve_alertsc Cst|j}|jd}|dkr(t|St|jdd}t|jdd}t|||}|j j ||j |d}t t |j}y&t||j j|||djdd Stjk r} zt|t| Sd} ~ XYnXdS) Nr2r3r4r5z-1)r(r7r8r)r9zapplication/json)rr(r.r:r;rr<r Zexecution_digest_run_tag_filterrr=r7r>r?r@r r rAr9rrBrrC) rrr0r2r3r5r8rDrErFrrrrs*   z(DebuggerV2Plugin.serve_execution_digestsc Cst|j}|jd}|dkr(t|St|jdd}t|jdd}t|||}|j j ||j |d}t t |j}y&t||j j|||djdd Stjk r} zt|t| Sd} ~ XYnXdS) Nr2r3r4r5z-1)r(r7r8r)r9zapplication/json)rr(r.r:r;rr<r Zexecution_data_run_tag_filterrr=r7r>r?r@r r rAr9rrBrrC) rrr0r2r3r5r8rDrErFrrrrs*   z%DebuggerV2Plugin.serve_execution_datac Cst|j}|jd}|dkr(t|St|jdd}t|jdd}t|||}|j j ||j |d}t t |j}y&t||j j|||djd d Stjk r} zt|t| Sd} ~ XYnXdS) aServe digests of intra-graph execution events. As the names imply, this route differs from `serve_execution_digests()` in that it is for intra-graph execution, while `serve_execution_digests()` is for top-level (eager) execution. r2Nr3r4r5z-1)r(r7r8r)r9zapplication/json)rr(r.r:r;rr<r Z%graph_execution_digest_run_tag_filterrr=r7r>r?r@r r rAr9rrBrrC) rrr0r2r3r5r8rDrErFrrrrs*   z.DebuggerV2Plugin.serve_graph_execution_digestsc Cst|j}|jd}|dkr(t|St|jdd}t|jdd}t|||}|j j ||j |d}t t |j}y&t||j j|||djd d Stjk r} zt|t| Sd} ~ XYnXdS) aServe detailed data objects of intra-graph execution events. As the names imply, this route differs from `serve_execution_data()` in that it is for intra-graph execution, while `serve_execution_data()` is for top-level (eager) execution. Unlike `serve_graph_execution_digests()`, this method serves the full-sized data objects for intra-graph execution events. r2Nr3r4r5z-1)r(r7r8r)r9zapplication/json)rr(r.r:r;rr<r Z#graph_execution_data_run_tag_filterrr=r7r>r?r@r r rAr9rrBrrC) rrr0r2r3r5r8rDrErFrrrr s*   z+DebuggerV2Plugin.serve_graph_execution_datac Cst|j}|jd}|dkr(t|S|jd}t||}|jj ||j |d}t t |j }y&t||jj|||djddStjk r}zt|t|Sd}~XYnXdS)aServe basic information about a TensorFlow graph. The request specifies the debugger-generated ID of the graph being queried. The response contains a JSON object with the following fields: - graph_id: The debugger-generated ID (echoing the request). - name: The name of the graph (if any). For TensorFlow 2.x Function Graphs (FuncGraphs), this is typically the name of the underlying Python function, optionally prefixed with TensorFlow-generated prefixed such as "__inference_". Some graphs (e.g., certain outermost graphs) may have no names, in which case this field is `null`. - outer_graph_id: Outer graph ID (if any). For an outermost graph without an outer graph context, this field is `null`. - inner_graph_ids: Debugger-generated IDs of all the graphs nested inside this graph. For a graph without any graphs nested inside, this field is an empty array. r2Ngraph_id)r(r7r8r)r9zapplication/json)rr(r.r:r;rr Zgraph_info_run_tag_filterrr=r7r>r?r@r r rAr9r NotFoundErrorrrC) rrr0r2rGr8rDrErFrrrr!s(   z!DebuggerV2Plugin.serve_graph_infoc Cst|j}|jd}|dkr(t|S|jd}|jd}t|||}|jj ||j |d}t t |j }y&t||jj|||djddStjk r} zt|t| Sd} ~ XYnXdS) a@ Serve information for ops in graphs. The request specifies the op name and the ID of the graph that contains the op. The response contains a JSON object with the following fields: - op_type - op_name - graph_ids: Stack of graph IDs that the op is located in, from outermost to innermost. The length of this array is always >= 1. The length is 1 if and only if the graph is an outermost graph. - num_outputs: Number of output tensors. - output_tensor_ids: The debugger-generated number IDs for the symbolic output tensors of the op (an array of numbers). - host_name: Name of the host on which the op is created. - stack_trace: Stack frames of the op's creation. - inputs: Specifications of all inputs to this op. Currently only immediate (one level of) inputs are provided. This is an array of length N_in, where N_in is the number of data inputs received by the op. Each element of the array is an object with the following fields: - op_name: Name of the op that provides the input tensor. - output_slot: 0-based output slot index from which the input tensor emits. - data: A recursive data structure of this same schema. This field is not populated (undefined) at the leaf nodes of this recursive data structure. In the rare case wherein the data for an input cannot be retrieved properly (e.g., special internal op types), this field will be unpopulated. This is an empty list for an op with no inputs. - consumers: Specifications for all the downstream consuming ops of this. Currently only immediate (one level of) consumers are provided. This is an array of length N_out, where N_out is the number of symbolic tensors output by this op. Each element of the array is an array of which the length equals the number of downstream ops that consume the corresponding symbolic tensor (only data edges are tracked). Each element of the array is an object with the following fields: - op_name: Name of the op that receives the output tensor as an input. - input_slot: 0-based input slot index at which the downstream op receives this output tensor. - data: A recursive data structure of this very schema. This field is not populated (undefined) at the leaf nodes of this recursive data structure. In the rare case wherein the data for a consumer op cannot be retrieved properly (e.g., special internal op types), this field will be unpopulated. If this op has no output tensors, this is an empty array. If one of the output tensors of this op has no consumers, the corresponding element is an empty array. r2NrGop_name)r(r7r8r)r9zapplication/json)rr(r.r:r;rr Zgraph_op_info_run_tag_filterrr=r7r>r?r@r r rAr9rrHrrC) rrr0r2rGrIr8rDrErFrrrr"/s*7     z$DebuggerV2Plugin.serve_graph_op_infocCszt|j}|jd}|dkr(t|St|}|jj ||j |d}t t |j }t||jj|||djddS)zCServes a list of all source files involved in the debugged program.r2N)r(r7r8r)r9zapplication/json)rr(r.r:r;rr Zsource_file_list_run_tag_filterrr=r7r>r?r@r r rAr9)rrr0r2r8rDrErrrr#s   z(DebuggerV2Plugin.serve_source_files_listc Cst|j}|jd}|dkr(t|S|jd}|dkrFt|dSt|}t ||}|j j ||j |d}t t|j}y&t||j j|||djddStjk r}zt|t|Sd}~XYnXdS) awServes the content of a given source file. The source file is referred to by the index in the list of all source files involved in the execution of the debugged program, which is available via the `serve_source_files_list()` serving route. Args: request: HTTP request. Returns: Response to the request. r2Nindexz-index is not provided for source file content)r(r7r8r)r9zapplication/json)rr(r.r:r;rrr<r Zsource_file_run_tag_filterrr=r7r>r?r@r r rAr9rrHrC) rrr0r2rJr8rDrErFrrrr$s0   z"DebuggerV2Plugin.serve_source_filec Cst|j}|jd}|dkr(t|S|jd}|dkrFt|dS|sTt|dS|d}t ||}|j j ||j |d}t t|j}y&t||j j|||djd d Stjk r}zt|t|Sd}~XYnXdS) aServes the content of stack frames. The source frames being requested are referred to be UUIDs for each of them, separated by commas. Args: request: HTTP request. Returns: Response to the request. r2Nstack_frame_idsz!Missing stack_frame_ids parameterzEmpty stack_frame_ids parameter,)r(r7r8r)r9zapplication/json)rr(r.r:r;rrsplitr Zstack_frames_run_tag_filterrr=r7r>r?r@r r rAr9rrHrC) rrr0r2rKr8rDrErFrrrr%s2      z#DebuggerV2Plugin.serve_stack_frames)__name__ __module__ __qualname____doc__r Z PLUGIN_NAMEr7rr&r+r,rRequestZ applicationrrrrrr r!r"r#r$r% __classcell__rr)rrr(s"   "%.Q.rN)rQ __future__rrrZwerkzeugrZ tensorboardrrZtensorboard.pluginsrZtensorboard.plugins.debugger_v2r Ztensorboard.backendr rrZTBPluginrrrrrs