retry rate limit bypass; new speakers

tts_service/deployed_models.toml

@@ -1,3 +1,17 @@
+[[speakers]]
+name = "Michael Rosen"
+slug = "michael-rosen"
+model_pipeline = "ArpabetGlowTtsMelgan"
+glow_tts = "glow-tts_michael-rosen_ljstx_2020.09.05_expr-1_chkpt-5682.torchjit"
+melgan = "melgan_manyvoice5.0_2020-07-23_12d5838_10760.torchjit"
+
+[[speakers]]
+name = "Wizard (Death Battle!)"
+slug = "wizard"
+model_pipeline = "ArpabetGlowTtsMelgan"
+glow_tts = "glow-tts_wiz_ljstx_2020.09.05_expr-1_chkpt-4500.torchjit"
+melgan = "melgan_wiz_2020.09.07_d3b9647_17000.torchjit"
+
 [[speakers]]
 name = "Tucker Carlson"
 slug = "tucker-carlson"

tts_service/models.toml

@@ -4,10 +4,24 @@ name = "Current Experiment"
 slug = "experiment"
 model_pipeline = "ArpabetGlowTtsMelgan"
 glow_tts = "glow_tts_single_speaker.torchjit"
-melgan = "melgan_manyvoice5.0_2020-07-23_12d5838_10760.torchjit"
+#melgan = "melgan_manyvoice5.0_2020-07-23_12d5838_10760.torchjit"
 #melgan = "melgan_manyvoice2.0_2020-07-08_12d5838_9330.torchjit"
 #melgan = "manyvoice1.0_2020-06-28_12d5838_6610.torchjit"
-#melgan = "melgan-vocodes.torchjit"
+melgan = "melgan-vocodes.torchjit"
+
+[[speakers]]
+name = "Michael Rosen"
+slug = "michael-rosen"
+model_pipeline = "ArpabetGlowTtsMelgan"
+glow_tts = "glow-tts_michael-rosen_ljstx_2020.09.05_expr-1_chkpt-5682.torchjit"
+melgan = "melgan_manyvoice5.0_2020-07-23_12d5838_10760.torchjit"
+
+[[speakers]]
+name = "Wizard (Death Battle!)"
+slug = "wizard"
+model_pipeline = "ArpabetGlowTtsMelgan"
+glow_tts = "glow-tts_wiz_ljstx_2020.09.05_expr-1_chkpt-4500.torchjit"
+melgan = "melgan_wiz_2020.09.07_d3b9647_17000.torchjit"
 
 [[speakers]]
 name = "Tucker Carlson"

tts_service/src/endpoints/speak/api.rs

@@ -8,8 +8,19 @@ use std::fmt;
 pub struct SpeakRequest {
   /// Slug for the speaker
   pub speaker: String,
+
   /// Raw text to be spoken
   pub text: String,
+
+  /// Retry attempt.
+  /// The first request is "0", and all subsequent requests are incremented.
+  /// Subsequent requests do not trigger the rate limiter.
+  #[serde(default)]
+  pub retry_attempt_number: u8,
+
+  /// Admin/debug requests get to skip the rate limiter.
+  #[serde(default)]
+  pub skip_rate_limiter: bool,
 }
 
 #[derive(Serialize,Debug)]

tts_service/src/endpoints/speak/speak.rs

@@ -35,7 +35,7 @@ pub async fn post_speak(request: HttpRequest,
 
   let ip_address = get_request_ip(&request);
 
-  if let Err(_err) = app_state.rate_limiter.maybe_ratelimit_request(&ip_address, &request.headers()) {
+  if let Err(_err) = app_state.rate_limiter.maybe_ratelimit_request(&ip_address, &request.headers(), &query) {
     return Err(SpeakError::rate_limited());
   }
 

tts_service/src/endpoints/speak/speak_with_spectrogram.rs

@@ -55,7 +55,7 @@ pub async fn post_speak_with_spectrogram(request: HttpRequest,
 
   let ip_address = get_request_ip(&request);
 
-  if let Err(_err) = app_state.rate_limiter.maybe_ratelimit_request(&ip_address, &request.headers()) {
+  if let Err(_err) = app_state.rate_limiter.maybe_ratelimit_request(&ip_address, &request.headers(), &query) {
     return Err(SpeakError::rate_limited());
   }
 

tts_service/src/rate_limiter/noop_rate_limiter.rs

@@ -1,11 +1,13 @@
 use crate::rate_limiter::rate_limiter::{RateLimiter, RateLimiterError};
 use actix_web::HttpRequest;
 use actix_web::http::HeaderMap;
+use crate::endpoints::speak::api::SpeakRequest;
 
 pub struct NoOpRateLimiter {}
 
 impl RateLimiter for NoOpRateLimiter {
-  fn maybe_ratelimit_request(&self, ip_address: &str, headers: &HeaderMap) -> Result<(), RateLimiterError> {
+  fn maybe_ratelimit_request(&self, ip_address: &str, headers: &HeaderMap,
+    speak_request: &SpeakRequest) -> Result<(), RateLimiterError> {
     Ok(())
   }
 

tts_service/src/rate_limiter/rate_limiter.rs

@@ -1,5 +1,6 @@
 use actix_web::HttpRequest;
 use actix_web::http::header::HeaderMap;
+use crate::endpoints::speak::api::SpeakRequest;
 
 #[derive(Debug)]
 pub enum RateLimiterError {
@@ -14,7 +15,8 @@ pub enum RateLimiterError {
 pub trait RateLimiter : Send + Sync {
   /// Potentially try to rate limit.
   /// Certain headers can be used to bypass.
-  fn maybe_ratelimit_request(&self, ip_address: &str, headers: &HeaderMap) -> Result<(), RateLimiterError>;
+  fn maybe_ratelimit_request(&self, ip_address: &str, headers: &HeaderMap,
+    speak_request: &SpeakRequest) -> Result<(), RateLimiterError>;
 
   /// Attempt to acquire the rate limiter.
   fn acquire(&self, rate_limit_key: &str) -> Result<(), RateLimiterError>;

tts_service/src/rate_limiter/redis_rate_limiter.rs

@@ -8,6 +8,7 @@ use actix_web::HttpRequest;
 use crate::endpoints::helpers::ip_address::get_request_ip;
 use actix_web::http::header::HeaderMap;
 use actix_web::http::HeaderName;
+use crate::endpoints::speak::api::SpeakRequest;
 
 const RATE_LIMIT_BYPASS_HEADER : &'static str = "limitless";
 
@@ -71,9 +72,19 @@ impl RedisRateLimiter {
 }
 
 impl RateLimiter for RedisRateLimiter {
-  fn maybe_ratelimit_request(&self, ip_address: &str, headers: &HeaderMap) -> Result<(), RateLimiterError> {
+  fn maybe_ratelimit_request(&self, ip_address: &str, headers: &HeaderMap, speak_request: &SpeakRequest) -> Result<(), RateLimiterError> {
+    if speak_request.skip_rate_limiter {
+      info!("Bypassing rate limiter with request param.");
+      return Ok(());
+    }
+
     if headers.contains_key(&self.rate_limit_bypass_header) {
-      info!("Bypassing rate limiter");
+      info!("Bypassing rate limiter with admin debug header.");
+      return Ok(());
+    }
+
+    if speak_request.retry_attempt_number > 0 {
+      info!("Retries bypass the rate limiter.");
       return Ok(());
     }