mirrors/twitter-algorithm
1
0
Fork 0
mirror of https://github.com/twitter/the-algorithm.git synced 2024-12-22 09:05:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

improvements from external prs

Browse source 
-fix corner case where dr converter failed when initializing

Closes twitter/the-algorithm#550
This commit is contained in:
twitter-team 2023-04-05 16:08:19 -07:00
parent 23fa75d406
commit 31e82d6474
31 changed files with 305 additions and 244 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
navi/dr_transform/src/all_config.rs
View file

@ -44,5 +44,6 @@ pub struct RenamedFeatures {
}
pub fn parse(json_str: &str) -> Result<AllConfig, Error> {
serde_json::from_str(json_str)
let all_config: AllConfig = serde_json::from_str(json_str)?;
Ok(all_config)
}

262
navi/dr_transform/src/converter.rs
View file

@ -2,6 +2,9 @@ use std::collections::BTreeSet;
use std::fmt::{self, Debug, Display};
use std::fs;
use crate::all_config;
use crate::all_config::AllConfig;
use anyhow::{bail, Context};
use bpr_thrift::data::DataRecord;
use bpr_thrift::prediction_service::BatchPredictionRequest;
use bpr_thrift::tensor::GeneralTensor;
@ -16,8 +19,6 @@ use segdense::util;
use thrift::protocol::{TBinaryInputProtocol, TSerializable};
use thrift::transport::TBufferChannel;
use crate::{all_config, all_config::AllConfig};
pub fn log_feature_match(
dr: &DataRecord,
seg_dense_config: &DensificationTransformSpec,
@ -28,20 +29,24 @@ pub fn log_feature_match(
for (feature_id, feature_value) in dr.continuous_features.as_ref().unwrap() {
debug!(
"{dr_type} - Continuous Datarecord => Feature ID: {feature_id}, Feature value: {feature_value}"
"{} - Continous Datarecord => Feature ID: {}, Feature value: {}",
dr_type, feature_id, feature_value
);
for input_feature in &seg_dense_config.cont.input_features {
if input_feature.feature_id == *feature_id {
debug!("Matching input feature: {input_feature:?}")
debug!("Matching input feature: {:?}", input_feature)
}
}
}
for feature_id in dr.binary_features.as_ref().unwrap() {
debug!("{dr_type} - Binary Datarecord => Feature ID: {feature_id}");
debug!(
"{} - Binary Datarecord => Feature ID: {}",
dr_type, feature_id
);
for input_feature in &seg_dense_config.binary.input_features {
if input_feature.feature_id == *feature_id {
debug!("Found input feature: {input_feature:?}")
debug!("Found input feature: {:?}", input_feature)
}
}
}
@ -90,18 +95,19 @@ impl BatchPredictionRequestToTorchTensorConverter {
model_version: &str,
reporting_feature_ids: Vec<(i64, &str)>,
register_metric_fn: Option<impl Fn(&HistogramVec)>,
) -> BatchPredictionRequestToTorchTensorConverter {
let all_config_path = format!("{model_dir}/{model_version}/all_config.json");
let seg_dense_config_path =
format!("{model_dir}/{model_version}/segdense_transform_spec_home_recap_2022.json");
let seg_dense_config = util::load_config(&seg_dense_config_path);
) -> anyhow::Result<BatchPredictionRequestToTorchTensorConverter> {
let all_config_path = format!("{}/{}/all_config.json", model_dir, model_version);
let seg_dense_config_path = format!(
"{}/{}/segdense_transform_spec_home_recap_2022.json",
model_dir, model_version
);
let seg_dense_config = util::load_config(&seg_dense_config_path)?;
let all_config = all_config::parse(
&fs::read_to_string(&all_config_path)
.unwrap_or_else(|error| panic!("error loading all_config.json - {error}")),
)
.unwrap();
.with_context(|| "error loading all_config.json - ")?,
)?;
let feature_mapper = util::load_from_parsed_config_ref(&seg_dense_config);
let feature_mapper = util::load_from_parsed_config(seg_dense_config.clone())?;
let user_embedding_feature_id = Self::get_feature_id(
&all_config
@ -131,11 +137,11 @@ impl BatchPredictionRequestToTorchTensorConverter {
let (discrete_feature_metrics, continuous_feature_metrics) = METRICS.get_or_init(|| {
let discrete = HistogramVec::new(
HistogramOpts::new(":navi:feature_id:discrete", "Discrete Feature ID values")
.buckets(Vec::from([
0.0f64, 10.0, 20.0, 30.0, 40.0, 50.0, 60.0, 70.0, 80.0, 90.0, 100.0, 110.0,
.buckets(Vec::from(&[
0.0, 10.0, 20.0, 30.0, 40.0, 50.0, 60.0, 70.0, 80.0, 90.0, 100.0, 110.0,
120.0, 130.0, 140.0, 150.0, 160.0, 170.0, 180.0, 190.0, 200.0, 250.0,
300.0, 500.0, 1000.0, 10000.0, 100000.0,
])),
] as &'static [f64])),
&["feature_id"],
)
.expect("metric cannot be created");
@ -144,18 +150,18 @@ impl BatchPredictionRequestToTorchTensorConverter {
":navi:feature_id:continuous",
"continuous Feature ID values",
)
.buckets(Vec::from([
0.0f64, 10.0, 20.0, 30.0, 40.0, 50.0, 60.0, 70.0, 80.0, 90.0, 100.0, 110.0,
120.0, 130.0, 140.0, 150.0, 160.0, 170.0, 180.0, 190.0, 200.0, 250.0, 300.0,
500.0, 1000.0, 10000.0, 100000.0,
])),
.buckets(Vec::from(&[
0.0, 10.0, 20.0, 30.0, 40.0, 50.0, 60.0, 70.0, 80.0, 90.0, 100.0, 110.0, 120.0,
130.0, 140.0, 150.0, 160.0, 170.0, 180.0, 190.0, 200.0, 250.0, 300.0, 500.0,
1000.0, 10000.0, 100000.0,
] as &'static [f64])),
&["feature_id"],
)
.expect("metric cannot be created");
if let Some(r) = register_metric_fn {
register_metric_fn.map(|r| {
r(&discrete);
r(&continuous);
}
});
(discrete, continuous)
});
@ -164,13 +170,16 @@ impl BatchPredictionRequestToTorchTensorConverter {
for (feature_id, feature_type) in reporting_feature_ids.iter() {
match *feature_type {
"discrete" => discrete_features_to_report.insert(*feature_id),
"continuous" => continuous_features_to_report.insert(*feature_id),
_ => panic!("Invalid feature type {feature_type} for reporting metrics!"),
"discrete" => discrete_features_to_report.insert(feature_id.clone()),
"continuous" => continuous_features_to_report.insert(feature_id.clone()),
_ => bail!(
"Invalid feature type {} for reporting metrics!",
feature_type
),
};
}
BatchPredictionRequestToTorchTensorConverter {
Ok(BatchPredictionRequestToTorchTensorConverter {
all_config,
seg_dense_config,
all_config_path,
@ -183,7 +192,7 @@ impl BatchPredictionRequestToTorchTensorConverter {
continuous_features_to_report,
discrete_feature_metrics,
continuous_feature_metrics,
}
})
}
fn get_feature_id(feature_name: &str, seg_dense_config: &Root) -> i64 {
@ -218,43 +227,45 @@ impl BatchPredictionRequestToTorchTensorConverter {
let mut working_set = vec![0 as f32; total_size];
let mut bpr_start = 0;
for (bpr, &bpr_end) in bprs.iter().zip(batch_size) {
if bpr.common_features.is_some()
&& bpr.common_features.as_ref().unwrap().tensors.is_some()
&& bpr
.common_features
.as_ref()
.unwrap()
.tensors
.as_ref()
.unwrap()
.contains_key(&feature_id)
{
let source_tensor = bpr
.common_features
.as_ref()
.unwrap()
.tensors
.as_ref()
.unwrap()
.get(&feature_id)
.unwrap();
let tensor = match source_tensor {
GeneralTensor::FloatTensor(float_tensor) =>
//Tensor::of_slice(
if bpr.common_features.is_some() {
if bpr.common_features.as_ref().unwrap().tensors.is_some() {
if bpr
.common_features
.as_ref()
.unwrap()
.tensors
.as_ref()
.unwrap()
.contains_key(&feature_id)
{
float_tensor
.floats
.iter()
.map(|x| x.into_inner() as f32)
.collect::<Vec<_>>()
}
_ => vec![0 as f32; cols],
};
let source_tensor = bpr
.common_features
.as_ref()
.unwrap()
.tensors
.as_ref()
.unwrap()
.get(&feature_id)
.unwrap();
let tensor = match source_tensor {
GeneralTensor::FloatTensor(float_tensor) =>
//Tensor::of_slice(
{
float_tensor
.floats
.iter()
.map(|x| x.into_inner() as f32)
.collect::<Vec<_>>()
}
_ => vec![0 as f32; cols],
};
// since the tensor is found in common feature, add it in all batches
for row in bpr_start..bpr_end {
for col in 0..cols {
working_set[row * cols + col] = tensor[col];
// since the tensor is found in common feature, add it in all batches
for row in bpr_start..bpr_end {
for col in 0..cols {
working_set[row * cols + col] = tensor[col];
}
}
}
}
}
@ -298,9 +309,9 @@ impl BatchPredictionRequestToTorchTensorConverter {
// (INT64 --> INT64, DataRecord.discrete_feature)
fn get_continuous(&self, bprs: &[BatchPredictionRequest], batch_ends: &[usize]) -> InputTensor {
// These need to be part of model schema
let rows = batch_ends[batch_ends.len() - 1];
let cols = 5293;
let full_size = rows * cols;
let rows: usize = batch_ends[batch_ends.len() - 1];
let cols: usize = 5293;
let full_size: usize = rows * cols;
let default_val = f32::NAN;
let mut tensor = vec![default_val; full_size];
@ -325,15 +336,18 @@ impl BatchPredictionRequestToTorchTensorConverter {
.unwrap();
for feature in common_features {
if let Some(f_info) = self.feature_mapper.get(feature.0) {
let idx = f_info.index_within_tensor as usize;
if idx < cols {
// Set value in each row
for r in bpr_start..bpr_end {
let flat_index = r * cols + idx;
tensor[flat_index] = feature.1.into_inner() as f32;
match self.feature_mapper.get(feature.0) {
Some(f_info) => {
let idx = f_info.index_within_tensor as usize;
if idx < cols {
// Set value in each row
for r in bpr_start..bpr_end {
let flat_index: usize = r * cols + idx;
tensor[flat_index] = feature.1.into_inner() as f32;
}
}
}
None => (),
}
if self.continuous_features_to_report.contains(feature.0) {
self.continuous_feature_metrics
@ -349,24 +363,28 @@ impl BatchPredictionRequestToTorchTensorConverter {
// Process the batch of datarecords
for r in bpr_start..bpr_end {
let dr: &DataRecord = &bpr.individual_features_list[r - bpr_start];
let dr: &DataRecord =
&bpr.individual_features_list[usize::try_from(r - bpr_start).unwrap()];
if dr.continuous_features.is_some() {
for feature in dr.continuous_features.as_ref().unwrap() {
if let Some(f_info) = self.feature_mapper.get(feature.0) {
let idx = f_info.index_within_tensor as usize;
let flat_index = r * cols + idx;
if flat_index < tensor.len() && idx < cols {
tensor[flat_index] = feature.1.into_inner() as f32;
match self.feature_mapper.get(&feature.0) {
Some(f_info) => {
let idx = f_info.index_within_tensor as usize;
let flat_index: usize = r * cols + idx;
if flat_index < tensor.len() && idx < cols {
tensor[flat_index] = feature.1.into_inner() as f32;
}
}
None => (),
}
if self.continuous_features_to_report.contains(feature.0) {
self.continuous_feature_metrics
.with_label_values(&[feature.0.to_string().as_str()])
.observe(feature.1.into_inner())
.observe(feature.1.into_inner() as f64)
} else if self.discrete_features_to_report.contains(feature.0) {
self.discrete_feature_metrics
.with_label_values(&[feature.0.to_string().as_str()])
.observe(feature.1.into_inner())
.observe(feature.1.into_inner() as f64)
}
}
}
@ -383,10 +401,10 @@ impl BatchPredictionRequestToTorchTensorConverter {
fn get_binary(&self, bprs: &[BatchPredictionRequest], batch_ends: &[usize]) -> InputTensor {
// These need to be part of model schema
let rows = batch_ends[batch_ends.len() - 1];
let cols = 149;
let full_size = rows * cols;
let default_val = 0;
let rows: usize = batch_ends[batch_ends.len() - 1];
let cols: usize = 149;
let full_size: usize = rows * cols;
let default_val: i64 = 0;
let mut v = vec![default_val; full_size];
@ -410,15 +428,18 @@ impl BatchPredictionRequestToTorchTensorConverter {
.unwrap();
for feature in common_features {
if let Some(f_info) = self.feature_mapper.get(feature) {
let idx = f_info.index_within_tensor as usize;
if idx < cols {
// Set value in each row
for r in bpr_start..bpr_end {
let flat_index = r * cols + idx;
v[flat_index] = 1;
match self.feature_mapper.get(feature) {
Some(f_info) => {
let idx = f_info.index_within_tensor as usize;
if idx < cols {
// Set value in each row
for r in bpr_start..bpr_end {
let flat_index: usize = r * cols + idx;
v[flat_index] = 1;
}
}
}
None => (),
}
}
}
@ -428,10 +449,13 @@ impl BatchPredictionRequestToTorchTensorConverter {
let dr: &DataRecord = &bpr.individual_features_list[r - bpr_start];
if dr.binary_features.is_some() {
for feature in dr.binary_features.as_ref().unwrap() {
if let Some(f_info) = self.feature_mapper.get(feature) {
let idx = f_info.index_within_tensor as usize;
let flat_index = r * cols + idx;
v[flat_index] = 1;
match self.feature_mapper.get(&feature) {
Some(f_info) => {
let idx = f_info.index_within_tensor as usize;
let flat_index: usize = r * cols + idx;
v[flat_index] = 1;
}
None => (),
}
}
}
@ -448,10 +472,10 @@ impl BatchPredictionRequestToTorchTensorConverter {
#[allow(dead_code)]
fn get_discrete(&self, bprs: &[BatchPredictionRequest], batch_ends: &[usize]) -> InputTensor {
// These need to be part of model schema
let rows = batch_ends[batch_ends.len() - 1];
let cols = 320;
let full_size = rows * cols;
let default_val = 0;
let rows: usize = batch_ends[batch_ends.len() - 1];
let cols: usize = 320;
let full_size: usize = rows * cols;
let default_val: i64 = 0;
let mut v = vec![default_val; full_size];
@ -475,15 +499,18 @@ impl BatchPredictionRequestToTorchTensorConverter {
.unwrap();
for feature in common_features {
if let Some(f_info) = self.feature_mapper.get(feature.0) {
let idx = f_info.index_within_tensor as usize;
if idx < cols {
// Set value in each row
for r in bpr_start..bpr_end {
let flat_index = r * cols + idx;
v[flat_index] = *feature.1;
match self.feature_mapper.get(feature.0) {
Some(f_info) => {
let idx = f_info.index_within_tensor as usize;
if idx < cols {
// Set value in each row
for r in bpr_start..bpr_end {
let flat_index: usize = r * cols + idx;
v[flat_index] = *feature.1;
}
}
}
None => (),
}
if self.discrete_features_to_report.contains(feature.0) {
self.discrete_feature_metrics
@ -495,15 +522,18 @@ impl BatchPredictionRequestToTorchTensorConverter {
// Process the batch of datarecords
for r in bpr_start..bpr_end {
let dr: &DataRecord = &bpr.individual_features_list[r];
let dr: &DataRecord = &bpr.individual_features_list[usize::try_from(r).unwrap()];
if dr.discrete_features.is_some() {
for feature in dr.discrete_features.as_ref().unwrap() {
if let Some(f_info) = self.feature_mapper.get(feature.0) {
let idx = f_info.index_within_tensor as usize;
let flat_index = r * cols + idx;
if flat_index < v.len() && idx < cols {
v[flat_index] = *feature.1;
match self.feature_mapper.get(&feature.0) {
Some(f_info) => {
let idx = f_info.index_within_tensor as usize;
let flat_index: usize = r * cols + idx;
if flat_index < v.len() && idx < cols {
v[flat_index] = *feature.1;
}
}
None => (),
}
if self.discrete_features_to_report.contains(feature.0) {
self.discrete_feature_metrics
@ -569,7 +599,7 @@ impl Converter for BatchPredictionRequestToTorchTensorConverter {
.map(|bpr| bpr.individual_features_list.len())
.scan(0usize, |acc, e| {
//running total
*acc += e;
*acc = *acc + e;
Some(*acc)
})
.collect::<Vec<_>>();

10
navi/navi/proto/tensorflow/core/framework/full_type.proto
View file

@ -122,7 +122,7 @@ enum FullTypeId {
// TFT_TENSOR[TFT_INT32, TFT_UNKNOWN]
// is a Tensor of int32 element type and unknown shape.
//
// TODO: Define TFT_SHAPE and add more examples.
// TODO(mdan): Define TFT_SHAPE and add more examples.
TFT_TENSOR = 1000;
// Array (or tensorflow::TensorList in the variant type registry).
@ -178,7 +178,7 @@ enum FullTypeId {
// object (for now).
// The bool element type.
// TODO
// TODO(mdan): Quantized types, legacy representations (e.g. ref)
TFT_BOOL = 200;
// Integer element types.
TFT_UINT8 = 201;
@ -195,7 +195,7 @@ enum FullTypeId {
TFT_DOUBLE = 211;
TFT_BFLOAT16 = 215;
// Complex element types.
// TODO: Represent as TFT_COMPLEX[TFT_DOUBLE] instead?
// TODO(mdan): Represent as TFT_COMPLEX[TFT_DOUBLE] instead?
TFT_COMPLEX64 = 212;
TFT_COMPLEX128 = 213;
// The string element type.
@ -240,7 +240,7 @@ enum FullTypeId {
// ownership is in the true sense: "the op argument representing the lock is
// available".
// Mutex locks are the dynamic counterpart of control dependencies.
// TODO: Properly document this thing.
// TODO(mdan): Properly document this thing.
//
// Parametrization: TFT_MUTEX_LOCK[].
TFT_MUTEX_LOCK = 10202;
@ -271,6 +271,6 @@ message FullTypeDef {
oneof attr {
string s = 3;
int64 i = 4;
// TODO: list/tensor, map? Need to reconcile with TFT_RECORD, etc.
// TODO(mdan): list/tensor, map? Need to reconcile with TFT_RECORD, etc.
}
}

2
navi/navi/proto/tensorflow/core/framework/function.proto
View file

@ -23,7 +23,7 @@ message FunctionDefLibrary {
// with a value. When a GraphDef has a call to a function, it must
// have binding for every attr defined in the signature.
//
// TODO:
// TODO(zhifengc):
// * device spec, etc.
message FunctionDef {
// The definition of the function's name, arguments, return values,

2
navi/navi/proto/tensorflow/core/framework/node_def.proto
View file

@ -61,7 +61,7 @@ message NodeDef {
// one of the names from the corresponding OpDef's attr field).
// The values must have a type matching the corresponding OpDef
// attr's type field.
// TODO: Add some examples here showing best practices.
// TODO(josh11b): Add some examples here showing best practices.
map<string, AttrValue> attr = 5;
message ExperimentalDebugInfo {

4
navi/navi/proto/tensorflow/core/framework/op_def.proto
View file

@ -96,7 +96,7 @@ message OpDef {
// Human-readable description.
string description = 4;
// TODO: bool is_optional?
// TODO(josh11b): bool is_optional?
// --- Constraints ---
// These constraints are only in effect if specified. Default is no
@ -139,7 +139,7 @@ message OpDef {
// taking input from multiple devices with a tree of aggregate ops
// that aggregate locally within each device (and possibly within
// groups of nearby devices) before communicating.
// TODO: Implement that optimization.
// TODO(josh11b): Implement that optimization.
bool is_aggregate = 16; // for things like add
// Other optimizations go here, like

2
navi/navi/proto/tensorflow/core/framework/step_stats.proto
View file

@ -53,7 +53,7 @@ message MemoryStats {
// Time/size stats recorded for a single execution of a graph node.
message NodeExecStats {
// TODO: Use some more compact form of node identity than
// TODO(tucker): Use some more compact form of node identity than
// the full string name. Either all processes should agree on a
// global id (cost_id?) for each node, or we should use a hash of
// the name.

2
navi/navi/proto/tensorflow/core/framework/tensor.proto
View file

@ -16,7 +16,7 @@ option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/framewo
message TensorProto {
DataType dtype = 1;
// Shape of the tensor. TODO: sort out the 0-rank issues.
// Shape of the tensor. TODO(touts): sort out the 0-rank issues.
TensorShapeProto tensor_shape = 2;
// Only one of the representations below is set, one of "tensor_contents" and

8
navi/navi/proto/tensorflow/core/protobuf/config.proto
View file

@ -532,7 +532,7 @@ message ConfigProto {
// We removed the flag client_handles_error_formatting. Marking the tag
// number as reserved.
// TODO: Should we just remove this tag so that it can be
// TODO(shikharagarwal): Should we just remove this tag so that it can be
// used in future for other purpose?
reserved 2;
@ -576,7 +576,7 @@ message ConfigProto {
// - If isolate_session_state is true, session states are isolated.
// - If isolate_session_state is false, session states are shared.
//
// TODO: Add a single API that consistently treats
// TODO(b/129330037): Add a single API that consistently treats
// isolate_session_state and ClusterSpec propagation.
bool share_session_state_in_clusterspec_propagation = 8;
@ -704,7 +704,7 @@ message ConfigProto {
// Options for a single Run() call.
message RunOptions {
// TODO Turn this into a TraceOptions proto which allows
// TODO(pbar) Turn this into a TraceOptions proto which allows
// tracing to be controlled in a more orthogonal manner?
enum TraceLevel {
NO_TRACE = 0;
@ -781,7 +781,7 @@ message RunMetadata {
repeated GraphDef partition_graphs = 3;
message FunctionGraphs {
// TODO: Include some sort of function/cache-key identifier?
// TODO(nareshmodi): Include some sort of function/cache-key identifier?
repeated GraphDef partition_graphs = 1;
GraphDef pre_optimization_graph = 2;

2
navi/navi/proto/tensorflow/core/protobuf/coordination_service.proto
View file

@ -194,7 +194,7 @@ service CoordinationService {
// Report error to the task. RPC sets the receiving instance of coordination
// service agent to error state permanently.
// TODO: Consider splitting this into a different RPC service.
// TODO(b/195990880): Consider splitting this into a different RPC service.
rpc ReportErrorToAgent(ReportErrorToAgentRequest)
returns (ReportErrorToAgentResponse);

2
navi/navi/proto/tensorflow/core/protobuf/debug.proto
View file

@ -46,7 +46,7 @@ message DebugTensorWatch {
// are to be debugged, the callers of Session::Run() must use distinct
// debug_urls to make sure that the streamed or dumped events do not overlap
// among the invocations.
// TODO: More visible documentation of this in g3docs.
// TODO(cais): More visible documentation of this in g3docs.
repeated string debug_urls = 4;
// Do not error out if debug op creation fails (e.g., due to dtype

6
navi/navi/proto/tensorflow/core/protobuf/debug_event.proto
View file

@ -12,7 +12,7 @@ option java_package = "org.tensorflow.util";
option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/protobuf/for_core_protos_go_proto";
// Available modes for extracting debugging information from a Tensor.
// TODO: Document the detailed column names and semantics in a separate
// TODO(cais): Document the detailed column names and semantics in a separate
// markdown file once the implementation settles.
enum TensorDebugMode {
UNSPECIFIED = 0;
@ -223,7 +223,7 @@ message DebuggedDevice {
// A debugger-generated ID for the device. Guaranteed to be unique within
// the scope of the debugged TensorFlow program, including single-host and
// multi-host settings.
// TODO: Test the uniqueness guarantee in multi-host settings.
// TODO(cais): Test the uniqueness guarantee in multi-host settings.
int32 device_id = 2;
}
@ -264,7 +264,7 @@ message Execution {
// field with the DebuggedDevice messages.
repeated int32 output_tensor_device_ids = 9;
// TODO support, add more fields
// TODO(cais): When backporting to V1 Session.run() support, add more fields
// such as fetches and feeds.
}

2
navi/navi/proto/tensorflow/core/protobuf/distributed_runtime_payloads.proto
View file

@ -7,7 +7,7 @@ option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/protobu
// Used to serialize and transmit tensorflow::Status payloads through
// grpc::Status `error_details` since grpc::Status lacks payload API.
// TODO: Use GRPC API once supported.
// TODO(b/204231601): Use GRPC API once supported.
message GrpcPayloadContainer {
map<string, bytes> payloads = 1;
}

2
navi/navi/proto/tensorflow/core/protobuf/eager_service.proto
View file

@ -172,7 +172,7 @@ message WaitQueueDoneRequest {
}
message WaitQueueDoneResponse {
// TODO: Consider adding NodeExecStats here to be able to
// TODO(nareshmodi): Consider adding NodeExecStats here to be able to
// propagate some stats.
}

2
navi/navi/proto/tensorflow/core/protobuf/master.proto
View file

@ -94,7 +94,7 @@ message ExtendSessionRequest {
}
message ExtendSessionResponse {
// TODO: Return something about the operation?
// TODO(mrry): Return something about the operation?
// The new version number for the extended graph, to be used in the next call
// to ExtendSession.

2
navi/navi/proto/tensorflow/core/protobuf/saved_object_graph.proto
View file

@ -176,7 +176,7 @@ message SavedBareConcreteFunction {
// allows the ConcreteFunction to be called with nest structure inputs. This
// field may not be populated. If this field is absent, the concrete function
// can only be called with flat inputs.
// TODO: support calling saved ConcreteFunction with structured
// TODO(b/169361281): support calling saved ConcreteFunction with structured
// inputs in C++ SavedModel API.
FunctionSpec function_spec = 4;
}

2
navi/navi/proto/tensorflow/core/protobuf/tensor_bundle.proto
View file

@ -17,7 +17,7 @@ option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/protobu
// Special header that is associated with a bundle.
//
// TODO: maybe in the future, we can add information about
// TODO(zongheng,zhifengc): maybe in the future, we can add information about
// which binary produced this checkpoint, timestamp, etc. Sometime, these can be
// valuable debugging information. And if needed, these can be used as defensive
// information ensuring reader (binary version) of the checkpoint and the writer

4
navi/navi/proto/tensorflow/core/protobuf/worker.proto
View file

@ -188,7 +188,7 @@ message DeregisterGraphRequest {
}
message DeregisterGraphResponse {
// TODO: Optionally add summary stats for the graph.
// TODO(mrry): Optionally add summary stats for the graph.
}
////////////////////////////////////////////////////////////////////////////////
@ -294,7 +294,7 @@ message RunGraphResponse {
// If the request asked for execution stats, the cost graph, or the partition
// graphs, these are returned here.
// TODO: Package these in a RunMetadata instead.
// TODO(suharshs): Package these in a RunMetadata instead.
StepStats step_stats = 2;
CostGraphDef cost_graph = 3;
repeated GraphDef partition_graph = 4;

2
navi/navi/proto/tensorflow_serving/apis/logging.proto
View file

@ -13,5 +13,5 @@ message LogMetadata {
SamplingConfig sampling_config = 2;
// List of tags used to load the relevant MetaGraphDef from SavedModel.
repeated string saved_model_tags = 3;
// TODO: Add more metadata as mentioned in the bug.
// TODO(b/33279154): Add more metadata as mentioned in the bug.
}

4
navi/navi/proto/tensorflow_serving/config/file_system_storage_path_source.proto
View file

@ -58,7 +58,7 @@ message FileSystemStoragePathSourceConfig {
// A single servable name/base_path pair to monitor.
// DEPRECATED: Use 'servables' instead.
// TODO: Stop using these fields, and ultimately remove them here.
// TODO(b/30898016): Stop using these fields, and ultimately remove them here.
string servable_name = 1 [deprecated = true];
string base_path = 2 [deprecated = true];
@ -76,7 +76,7 @@ message FileSystemStoragePathSourceConfig {
// check for a version to appear later.)
// DEPRECATED: Use 'servable_versions_always_present' instead, which includes
// this behavior.
// TODO: Remove 2019-10-31 or later.
// TODO(b/30898016): Remove 2019-10-31 or later.
bool fail_if_zero_versions_at_startup = 4 [deprecated = true];
// If true, the servable is always expected to exist on the underlying

4
navi/navi/proto/tensorflow_serving/config/model_server_config.proto
View file

@ -9,7 +9,7 @@ import "tensorflow_serving/config/logging_config.proto";
option cc_enable_arenas = true;
// The type of model.
// TODO: DEPRECATED.
// TODO(b/31336131): DEPRECATED.
enum ModelType {
MODEL_TYPE_UNSPECIFIED = 0 [deprecated = true];
TENSORFLOW = 1 [deprecated = true];
@ -31,7 +31,7 @@ message ModelConfig {
string base_path = 2;
// Type of model.
// TODO: DEPRECATED. Please use 'model_platform' instead.
// TODO(b/31336131): DEPRECATED. Please use 'model_platform' instead.
ModelType model_type = 3 [deprecated = true];
// Type of model (e.g. "tensorflow").

24
navi/navi/src/bootstrap.rs
View file

@ -1,5 +1,6 @@
use anyhow::Result;
use log::{info, warn};
use x509_parser::{prelude::{parse_x509_pem}, parse_x509_certificate};
use std::collections::HashMap;
use tokio::time::Instant;
use tonic::{
@ -27,6 +28,7 @@ use crate::cli_args::{ARGS, INPUTS, OUTPUTS};
use crate::metrics::{
NAVI_VERSION, NUM_PREDICTIONS, NUM_REQUESTS_FAILED, NUM_REQUESTS_FAILED_BY_MODEL,
NUM_REQUESTS_RECEIVED, NUM_REQUESTS_RECEIVED_BY_MODEL, RESPONSE_TIME_COLLECTOR,
CERT_EXPIRY_EPOCH
};
use crate::predict_service::{Model, PredictService};
use crate::tf_proto::tensorflow_serving::model_spec::VersionChoice::Version;
@ -233,6 +235,12 @@ impl<T: Model> PredictionService for PredictService<T> {
}
}
// A function that takes a timestamp as input and returns a ticker stream
fn report_expiry(expiry_time: i64) {
info!("Certificate expires at epoch: {:?}", expiry_time);
CERT_EXPIRY_EPOCH.set(expiry_time as i64);
}
pub fn bootstrap<T: Model>(model_factory: ModelFactory<T>) -> Result<()> {
info!("package: {}, version: {}, args: {:?}", NAME, VERSION, *ARGS);
//we follow SemVer. So here we assume MAJOR.MINOR.PATCH
@ -249,6 +257,7 @@ pub fn bootstrap<T: Model>(model_factory: ModelFactory<T>) -> Result<()> {
);
}
tokio::runtime::Builder::new_multi_thread()
.thread_name("async worker")
.worker_threads(ARGS.num_worker_threads)
@ -266,6 +275,21 @@ pub fn bootstrap<T: Model>(model_factory: ModelFactory<T>) -> Result<()> {
let mut builder = if ARGS.ssl_dir.is_empty() {
Server::builder()
} else {
// Read the pem file as a string
let pem_str = std::fs::read_to_string(format!("{}/server.crt", ARGS.ssl_dir)).unwrap();
let res = parse_x509_pem(&pem_str.as_bytes());
match res {
Ok((rem, pem_2)) => {
assert!(rem.is_empty());
assert_eq!(pem_2.label, String::from("CERTIFICATE"));
let res_x509 = parse_x509_certificate(&pem_2.contents);
info!("Certificate label: {}", pem_2.label);
assert!(res_x509.is_ok());
report_expiry(res_x509.unwrap().1.validity().not_after.timestamp());
},
_ => panic!("PEM parsing failed: {:?}", res),
}
let key = tokio::fs::read(format!("{}/server.key", ARGS.ssl_dir))
.await
.expect("can't find key file");

7
navi/navi/src/metrics.rs
View file

@ -171,6 +171,9 @@ lazy_static! {
&["model_name"]
)
.expect("metric can be created");
pub static ref CERT_EXPIRY_EPOCH: IntGauge =
IntGauge::new(":navi:cert_expiry_epoch", "Timestamp when the current cert expires")
.expect("metric can be created");
}
pub fn register_custom_metrics() {
@ -249,6 +252,10 @@ pub fn register_custom_metrics() {
REGISTRY
.register(Box::new(CONVERTER_TIME_COLLECTOR.clone()))
.expect("collector can be registered");
REGISTRY
.register(Box::new(CERT_EXPIRY_EPOCH.clone()))
.expect("collector can be registered");
}
pub fn register_dynamic_metrics(c: &HistogramVec) {

2
navi/navi/src/onnx_model.rs
View file

@ -189,7 +189,7 @@ pub mod onnx {
&version,
reporting_feature_ids,
Some(metrics::register_dynamic_metrics),
)),
)?),
};
onnx_model.warmup()?;
Ok(onnx_model)

40
navi/navi/src/predict_service.rs
View file

@ -24,7 +24,7 @@ use serde_json::{self, Value};
pub trait Model: Send + Sync + Display + Debug + 'static {
fn warmup(&self) -> Result<()>;
//TODO: refactor this to return Vec<Vec<TensorScores>>, i.e.
//TODO: refactor this to return vec<vec<TensorScores>>, i.e.
//we have the underlying runtime impl to split the response to each client.
//It will eliminate some inefficient memory copy in onnx_model.rs as well as simplify code
fn do_predict(
@ -222,8 +222,8 @@ impl<T: Model> PredictService<T> {
.map(|b| b.parse().unwrap())
.collect::<Vec<u64>>();
let no_msg_wait_millis = *batch_time_out_millis.iter().min().unwrap();
let mut all_model_predictors =
ArrayVec::<ArrayVec<BatchPredictor<T>, MAX_VERSIONS_PER_MODEL>, MAX_NUM_MODELS>::new();
let mut all_model_predictors: ArrayVec::<ArrayVec<BatchPredictor<T>, MAX_VERSIONS_PER_MODEL>, MAX_NUM_MODELS> =
(0 ..MAX_NUM_MODELS).map( |_| ArrayVec::<BatchPredictor<T>, MAX_VERSIONS_PER_MODEL>::new()).collect();
loop {
let msg = rx.try_recv();
let no_more_msg = match msg {
@ -272,27 +272,23 @@ impl<T: Model> PredictService<T> {
queue_reset_ts: Instant::now(),
queue_earliest_rq_ts: Instant::now(),
};
if idx < all_model_predictors.len() {
metrics::NEW_MODEL_SNAPSHOT
.with_label_values(&[&MODEL_SPECS[idx]])
.inc();
assert!(idx < all_model_predictors.len());
metrics::NEW_MODEL_SNAPSHOT
.with_label_values(&[&MODEL_SPECS[idx]])
.inc();
info!("now we serve updated model: {}", predictor.model);
//we can do this since the vector is small
let predictors = &mut all_model_predictors[idx];
if predictors.len() == ARGS.versions_per_model {
predictors.remove(predictors.len() - 1);
}
predictors.insert(0, predictor);
} else {
info!("now we serve new model: {:}", predictor.model);
let mut predictors =
ArrayVec::<BatchPredictor<T>, MAX_VERSIONS_PER_MODEL>::new();
predictors.push(predictor);
all_model_predictors.push(predictors);
//check the invariant that we always push the last model to the end
assert_eq!(all_model_predictors.len(), idx + 1)
//we can do this since the vector is small
let predictors = &mut all_model_predictors[idx];
if predictors.len() == 0 {
info!("now we serve new model: {}", predictor.model);
}
else {
info!("now we serve updated model: {}", predictor.model);
}
if predictors.len() == ARGS.versions_per_model {
predictors.remove(predictors.len() - 1);
}
predictors.insert(0, predictor);
false
}
Err(TryRecvError::Empty) => true,

56
navi/segdense/src/error.rs
View file

@ -5,39 +5,49 @@ use std::fmt::Display;
*/
#[derive(Debug)]
pub enum SegDenseError {
IoError(std::io::Error),
Json(serde_json::Error),
JsonMissingRoot,
JsonMissingObject,
JsonMissingArray,
JsonArraySize,
JsonMissingInputFeature,
IoError(std::io::Error),
Json(serde_json::Error),
JsonMissingRoot,
JsonMissingObject,
JsonMissingArray,
JsonArraySize,
JsonMissingInputFeature,
}
impl Display for SegDenseError {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
match self {
SegDenseError::IoError(io_error) => write!(f, "{}", io_error),
SegDenseError::Json(serde_json) => write!(f, "{}", serde_json),
SegDenseError::JsonMissingRoot => write!(f, "{}", "SegDense JSON: Root Node note found!"),
SegDenseError::JsonMissingObject => write!(f, "{}", "SegDense JSON: Object note found!"),
SegDenseError::JsonMissingArray => write!(f, "{}", "SegDense JSON: Array Node note found!"),
SegDenseError::JsonArraySize => write!(f, "{}", "SegDense JSON: Array size not as expected!"),
SegDenseError::JsonMissingInputFeature => write!(f, "{}", "SegDense JSON: Missing input feature!"),
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
match self {
SegDenseError::IoError(io_error) => write!(f, "{}", io_error),
SegDenseError::Json(serde_json) => write!(f, "{}", serde_json),
SegDenseError::JsonMissingRoot => {
write!(f, "{}", "SegDense JSON: Root Node note found!")
}
SegDenseError::JsonMissingObject => {
write!(f, "{}", "SegDense JSON: Object note found!")
}
SegDenseError::JsonMissingArray => {
write!(f, "{}", "SegDense JSON: Array Node note found!")
}
SegDenseError::JsonArraySize => {
write!(f, "{}", "SegDense JSON: Array size not as expected!")
}
SegDenseError::JsonMissingInputFeature => {
write!(f, "{}", "SegDense JSON: Missing input feature!")
}
}
}
}
}
impl std::error::Error for SegDenseError {}
impl From<std::io::Error> for SegDenseError {
fn from(err: std::io::Error) -> Self {
SegDenseError::IoError(err)
}
fn from(err: std::io::Error) -> Self {
SegDenseError::IoError(err)
}
}
impl From<serde_json::Error> for SegDenseError {
fn from(err: serde_json::Error) -> Self {
SegDenseError::Json(err)
}
fn from(err: serde_json::Error) -> Self {
SegDenseError::Json(err)
}
}

2
navi/segdense/src/lib.rs
View file

@ -1,4 +1,4 @@
pub mod error;
pub mod segdense_transform_spec_home_recap_2022;
pub mod mapper;
pub mod segdense_transform_spec_home_recap_2022;
pub mod util;

21
navi/segdense/src/main.rs
View file

@ -5,19 +5,18 @@ use segdense::error::SegDenseError;
use segdense::util;
fn main() -> Result<(), SegDenseError> {
env_logger::init();
let args: Vec<String> = env::args().collect();
env_logger::init();
let args: Vec<String> = env::args().collect();
let schema_file_name: &str = if args.len() == 1 {
"json/compact.json"
} else {
&args[1]
};
let schema_file_name: &str = if args.len() == 1 {
"json/compact.json"
} else {
&args[1]
};
let json_str = fs::read_to_string(schema_file_name)?;
let json_str = fs::read_to_string(schema_file_name)?;
util::safe_load_config(&json_str)?;
util::safe_load_config(&json_str)?;
Ok(())
Ok(())
}

2
navi/segdense/src/mapper.rs
View file

@ -19,7 +19,7 @@ pub struct FeatureMapper {
impl FeatureMapper {
pub fn new() -> FeatureMapper {
FeatureMapper {
map: HashMap::new()
map: HashMap::new(),
}
}
}

1
navi/segdense/src/segdense_transform_spec_home_recap_2022.rs
View file

@ -164,7 +164,6 @@ pub struct ComplexFeatureTypeTransformSpec {
pub tensor_shape: Vec<i64>,
}
#[derive(Default, Debug, Clone, PartialEq, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct InputFeatureMapRecord {

55
navi/segdense/src/util.rs
View file

@ -1,23 +1,23 @@
use log::debug;
use std::fs;
use log::{debug};
use serde_json::{Value, Map};
use serde_json::{Map, Value};
use crate::error::SegDenseError;
use crate::mapper::{FeatureMapper, FeatureInfo, MapWriter};
use crate::mapper::{FeatureInfo, FeatureMapper, MapWriter};
use crate::segdense_transform_spec_home_recap_2022::{self as seg_dense, InputFeature};
pub fn load_config(file_name: &str) -> seg_dense::Root {
let json_str = fs::read_to_string(file_name).expect(
&format!("Unable to load segdense file {}", file_name));
let seg_dense_config = parse(&json_str).expect(
&format!("Unable to parse segdense file {}", file_name));
return seg_dense_config;
pub fn load_config(file_name: &str) -> Result<seg_dense::Root, SegDenseError> {
let json_str = fs::read_to_string(file_name)?;
// &format!("Unable to load segdense file {}", file_name));
let seg_dense_config = parse(&json_str)?;
// &format!("Unable to parse segdense file {}", file_name));
Ok(seg_dense_config)
}
pub fn parse(json_str: &str) -> Result<seg_dense::Root, SegDenseError> {
let root: seg_dense::Root = serde_json::from_str(json_str)?;
return Ok(root);
Ok(root)
}
/**
@ -44,15 +44,8 @@ pub fn safe_load_config(json_str: &str) -> Result<FeatureMapper, SegDenseError>
load_from_parsed_config(root)
}
pub fn load_from_parsed_config_ref(root: &seg_dense::Root) -> FeatureMapper {
load_from_parsed_config(root.clone()).unwrap_or_else(
|error| panic!("Error loading all_config.json - {}", error))
}
// Perf note : make 'root' un-owned
pub fn load_from_parsed_config(root: seg_dense::Root) ->
Result<FeatureMapper, SegDenseError> {
pub fn load_from_parsed_config(root: seg_dense::Root) -> Result<FeatureMapper, SegDenseError> {
let v = root.input_features_map;
// Do error check
@ -86,7 +79,7 @@ pub fn load_from_parsed_config(root: seg_dense::Root) ->
Some(info) => {
debug!("{:?}", info);
fm.set(feature_id, info)
},
}
None => (),
}
}
@ -94,19 +87,22 @@ pub fn load_from_parsed_config(root: seg_dense::Root) ->
Ok(fm)
}
#[allow(dead_code)]
fn add_feature_info_to_mapper(feature_mapper: &mut FeatureMapper, input_features: &Vec<InputFeature>) {
fn add_feature_info_to_mapper(
feature_mapper: &mut FeatureMapper,
input_features: &Vec<InputFeature>,
) {
for input_feature in input_features.iter() {
let feature_id = input_feature.feature_id;
let feature_info = to_feature_info(input_feature);
let feature_id = input_feature.feature_id;
let feature_info = to_feature_info(input_feature);
match feature_info {
Some(info) => {
debug!("{:?}", info);
feature_mapper.set(feature_id, info)
},
None => (),
match feature_info {
Some(info) => {
debug!("{:?}", info);
feature_mapper.set(feature_id, info)
}
None => (),
}
}
}
pub fn to_feature_info(input_feature: &seg_dense::InputFeature) -> Option<FeatureInfo> {
@ -139,7 +135,7 @@ pub fn to_feature_info(input_feature: &seg_dense::InputFeature) -> Option<Featur
2 => 0,
3 => 2,
_ => -1,
}
},
};
if input_feature.index < 0 {
@ -156,4 +152,3 @@ pub fn to_feature_info(input_feature: &seg_dense::InputFeature) -> Option<Featur
index_within_tensor: input_feature.index,
})
}