feat: add healthchecks (#93)

* feat: add healthchecks * fix: six version
deeppavlov · Oct 24, 2023 · d9ff186 · d9ff186
1 parent 8432f1e
commit d9ff186
Show file tree

Hide file tree

Showing 87 changed files with 190 additions and 61 deletions.
diff --git a/annotators/IntentCatcherTransformers/requirements.txt b/annotators/IntentCatcherTransformers/requirements.txt
@@ -5,6 +5,7 @@ itsdangerous==2.0.1
 gunicorn==19.9.0
 requests==2.22.0
 sentry-sdk[flask]==0.14.1
+six==1.16.0
 healthcheck==1.3.3
 jinja2<=3.0.3
 Werkzeug<=2.0.3

diff --git a/annotators/MultilingualSentimentClassification/requirements.txt b/annotators/MultilingualSentimentClassification/requirements.txt
@@ -3,6 +3,7 @@ flask==1.1.1
 gunicorn==19.9.0
 requests==2.22.0
 sentry-sdk[flask]==0.14.1
+six==1.16.0
 healthcheck==1.3.3
 itsdangerous==2.0.1
 jinja2<=3.0.3

diff --git a/annotators/MultilingualToxicClassification/requirements.txt b/annotators/MultilingualToxicClassification/requirements.txt
@@ -4,6 +4,7 @@ flask==1.1.1
 gunicorn==19.9.0
 requests==2.22.0
 sentry-sdk[flask]==0.14.1
+six==1.16.0
 healthcheck==1.3.3
 itsdangerous==2.0.1
 jinja2<=3.0.3

diff --git a/annotators/combined_classification/Dockerfile b/annotators/combined_classification/Dockerfile
@@ -19,4 +19,7 @@ ENV KEEP_ZERO_CONFS=$KEEP_ZERO_CONFS
 COPY annotators/combined_classification/ ./
 COPY common/ common/
 
+# wait for a server answer ( INTERVAL + TIMEOUT ) * RETRIES seconds after that change stutus to unhealthy
+HEALTHCHECK --interval=5s --timeout=5s --retries=3 CMD curl --fail 127.0.0.1:${SERVICE_PORT}/healthcheck || exit 1
+
 CMD gunicorn --workers=1 server:app -b 0.0.0.0:${SERVICE_PORT} --timeout=1200 --preload
diff --git a/annotators/combined_classification/requirements.txt b/annotators/combined_classification/requirements.txt
@@ -8,3 +8,5 @@ transformers==4.15.0
 jinja2<=3.0.3
 Werkzeug<=2.0.3
 pytorch-crf==0.7.2
+six==1.16.0
+healthcheck==1.3.3
diff --git a/annotators/combined_classification/server.py b/annotators/combined_classification/server.py
@@ -2,9 +2,9 @@
 import os
 import time
 
-from flask import Flask, request, jsonify
 import sentry_sdk
-
+from flask import Flask, request, jsonify
+from healthcheck import HealthCheck
 from sentry_sdk.integrations.flask import FlaskIntegration
 from deeppavlov import build_model
 from common.utils import combined_classes
@@ -66,6 +66,7 @@ def get_result(sentences, sentences_with_history, postannotations=False):
     raise e
 
 app = Flask(__name__)
+health = HealthCheck(app, "/healthcheck")
 
 
 @app.route("/model", methods=["POST"])

diff --git a/annotators/doc_processor/Dockerfile b/annotators/doc_processor/Dockerfile
@@ -18,4 +18,7 @@ COPY annotators/doc_processor /src
 COPY documents /src/documents
 COPY common /src/common
 
+# wait for a server answer ( INTERVAL + TIMEOUT ) * RETRIES seconds after that change stutus to unhealthy
+HEALTHCHECK --interval=5s --timeout=5s --retries=3 CMD curl --fail 127.0.0.1:${SERVICE_PORT}/healthcheck || exit 1
+
 CMD gunicorn --workers=1 server:app -b 0.0.0.0:${SERVICE_PORT} --timeout=1200
diff --git a/annotators/doc_processor/requirements.txt b/annotators/doc_processor/requirements.txt
@@ -9,4 +9,6 @@ pypdfium2==4.16.0
 bs4==0.0.1
 python-docx==0.8.11
 simplify-docx==0.1.2
-filetype==1.2.0
+filetype==1.2.0
+six==1.16.0
+healthcheck==1.3.3
diff --git a/annotators/doc_processor/server.py b/annotators/doc_processor/server.py
@@ -1,17 +1,20 @@
 import logging
 import os
+
 import sentry_sdk
+from common.files_and_folders_processing import SKILLS_USING_DOC
 from flask import Flask, jsonify, request
+from healthcheck import HealthCheck
 from sentry_sdk.integrations.flask import FlaskIntegration
 from utils import upload_documents_save_info
-from common.files_and_folders_processing import SKILLS_USING_DOC
 
 # logging here because it conflicts with tf
 
 logging.basicConfig(format="%(asctime)s - %(name)s - %(levelname)s - %(message)s", level=logging.INFO)
 logger = logging.getLogger(__name__)
 sentry_sdk.init(dsn=os.getenv("SENTRY_DSN"), integrations=[FlaskIntegration()])
 app = Flask(__name__)
+health = HealthCheck(app, "/healthcheck")
 
 DOC_PATHS_OR_LINKS = os.environ.get("DOC_PATHS_OR_LINKS")
 if DOC_PATHS_OR_LINKS:

diff --git a/annotators/doc_retriever/Dockerfile b/annotators/doc_retriever/Dockerfile
@@ -18,4 +18,7 @@ COPY annotators/doc_retriever /src
 COPY documents /src/documents
 COPY common /src/common
 
+# wait for a server answer ( INTERVAL + TIMEOUT ) * RETRIES seconds after that change stutus to unhealthy
+HEALTHCHECK --interval=5s --timeout=5s --retries=3 CMD curl --fail 127.0.0.1:${SERVICE_PORT}/healthcheck || exit 1
+
 CMD gunicorn --workers=1 server:app -b 0.0.0.0:${SERVICE_PORT} --timeout=1200
diff --git a/annotators/doc_retriever/requirements.txt b/annotators/doc_retriever/requirements.txt
@@ -3,6 +3,7 @@ itsdangerous==2.0.1
 gunicorn==19.9.0
 requests==2.22.0
 sentry-sdk[flask]==0.14.1
+six==1.16.0
 healthcheck==1.3.3
 jinja2<=3.0.3
 Werkzeug<=2.0.3
@@ -12,4 +13,6 @@ spacy==3.5
 pypdfium2==4.16.0
 bs4==0.0.1
 python-docx==0.8.11
-simplify-docx==0.1.2
+simplify-docx==0.1.2
+six==1.16.0
+healthcheck==1.3.3
diff --git a/annotators/doc_retriever/server.py b/annotators/doc_retriever/server.py
@@ -1,21 +1,24 @@
 import logging
 import os
-import sentry_sdk
-import requests
 from pathlib import PurePath
+
+import requests
+import sentry_sdk
+from common.files_and_folders_processing import create_folders_if_not_exist
 from deeppavlov import build_model
+from deeppavlov.core.common.file import read_json
 from flask import Flask, jsonify, request
+from healthcheck import HealthCheck
 from sentry_sdk.integrations.flask import FlaskIntegration
-from deeppavlov.core.common.file import read_json
 from utils import vectorize_upload_return_attributes, download_files, add_file_id_to_config
-from common.files_and_folders_processing import create_folders_if_not_exist
 
 # logging here because it conflicts with tf
 
 logging.basicConfig(format="%(asctime)s - %(name)s - %(levelname)s - %(message)s", level=logging.INFO)
 logger = logging.getLogger(__name__)
 sentry_sdk.init(dsn=os.getenv("SENTRY_DSN"), integrations=[FlaskIntegration()])
 app = Flask(__name__)
+health = HealthCheck(app, "/healthcheck")
 
 PARAGRAPHS_NUM = int(os.environ.get("PARAGRAPHS_NUM", 5))
 FILE_SERVER_TIMEOUT = float(os.environ.get("FILE_SERVER_TIMEOUT", 30))

diff --git a/annotators/entity_storer/requirements.txt b/annotators/entity_storer/requirements.txt
@@ -2,6 +2,7 @@ sentry_sdk[flask]==1.19.1
 flask==2.2.3
 itsdangerous==2.0.1
 gunicorn==20.1.0
+six==1.16.0
 healthcheck==1.3.3
 six==1.15.0
 spacy==3.5.1

diff --git a/annotators/prompt_selector/Dockerfile b/annotators/prompt_selector/Dockerfile
@@ -15,4 +15,7 @@ ENV SENTENCE_RANKER_SERVICE_URL ${SENTENCE_RANKER_SERVICE_URL}
 ARG N_SENTENCES_TO_RETURN
 ENV N_SENTENCES_TO_RETURN ${N_SENTENCES_TO_RETURN}
 
+# wait for a server answer ( INTERVAL + TIMEOUT ) * RETRIES seconds after that change stutus to unhealthy
+HEALTHCHECK --interval=5s --timeout=5s --retries=3 CMD curl --fail 127.0.0.1:${SERVICE_PORT}/healthcheck || exit 1
+
 CMD gunicorn --workers=1 server:app --bind 0.0.0.0:8000
diff --git a/annotators/prompt_selector/requirements.txt b/annotators/prompt_selector/requirements.txt
@@ -7,3 +7,5 @@ click<=8.0.4
 jinja2<=3.1.2
 Werkzeug>=2.2.2,<3.0
 numpy>=1.17.2
+six==1.16.0
+healthcheck==1.3.3
diff --git a/annotators/prompt_selector/server.py b/annotators/prompt_selector/server.py
@@ -10,12 +10,14 @@
 import numpy as np
 import sentry_sdk
 from flask import Flask, request, jsonify
+from healthcheck import HealthCheck
 
 
 sentry_sdk.init(getenv("SENTRY_DSN"))
 logging.basicConfig(format="%(asctime)s - %(name)s - %(levelname)s - %(message)s", level=logging.DEBUG)
 logger = logging.getLogger(__name__)
 app = Flask(__name__)
+health = HealthCheck(app, "/healthcheck")
 
 SENTENCE_RANKER_SERVICE_URL = getenv("SENTENCE_RANKER_SERVICE_URL")
 logger.info(f"prompt-selector considered ranker: {SENTENCE_RANKER_SERVICE_URL}")

diff --git a/annotators/prompt_selector/template.yml b/annotators/prompt_selector/template.yml
diff --git a/annotators/toxic_classification_ru/requirements.txt b/annotators/toxic_classification_ru/requirements.txt
@@ -4,6 +4,7 @@ flask==1.1.1
 gunicorn==19.9.0
 requests==2.22.0
 sentry-sdk[flask]==0.14.1
+six==1.16.0
 healthcheck==1.3.3
 itsdangerous==2.0.1
 jinja2<=3.0.3

diff --git a/common/dialogflow_framework/requirements.txt b/common/dialogflow_framework/requirements.txt
@@ -3,6 +3,7 @@ sentry_sdk[flask]==0.14.3
 flask==1.1.1
 itsdangerous==2.0.1
 gunicorn==19.9.0
+six==1.16.0
 healthcheck==1.3.3
 # dialogflow framework
 programy==4.3

diff --git a/response_selectors/confidence_based_response_selector/Dockerfile b/response_selectors/confidence_based_response_selector/Dockerfile
@@ -12,4 +12,7 @@ ENV FILTER_BADLISTED_WORDS ${FILTER_BADLISTED_WORDS}
 
 COPY . /src/
 
+# wait for a server answer ( INTERVAL + TIMEOUT ) * RETRIES seconds after that change stutus to unhealthy
+HEALTHCHECK --interval=5s --timeout=5s --retries=3 CMD curl --fail 127.0.0.1:${SERVICE_PORT}/healthcheck || exit 1
+
 CMD gunicorn --workers=1 --bind 0.0.0.0:3003 server:app
diff --git a/response_selectors/confidence_based_response_selector/component.yml b/response_selectors/confidence_based_response_selector/component.yml
diff --git a/response_selectors/confidence_based_response_selector/pipeline.yml b/response_selectors/confidence_based_response_selector/pipeline.yml
diff --git a/response_selectors/confidence_based_response_selector/requirements.txt b/response_selectors/confidence_based_response_selector/requirements.txt
@@ -6,3 +6,5 @@ numpy==1.25.0
 sentry-sdk==0.12.3
 jinja2<=3.0.3
 Werkzeug<=2.0.3
+six==1.16.0
+healthcheck==1.3.3
diff --git a/response_selectors/confidence_based_response_selector/server.py b/response_selectors/confidence_based_response_selector/server.py
@@ -5,9 +5,10 @@
 import time
 import requests
 
+import sentry_sdk
 from flask import Flask, request, jsonify
+from healthcheck import HealthCheck
 from os import getenv
-import sentry_sdk
 
 
 sentry_sdk.init(getenv("SENTRY_DSN"))
@@ -16,6 +17,7 @@
 logger = logging.getLogger(__name__)
 
 app = Flask(__name__)
+health = HealthCheck(app, "/healthcheck")
 
 BADLIST_URL = getenv("BADLIST_ANNOTATOR_URL", "http://badlisted-words:8018/badlisted_words_batch")
 FILTER_BADLISTED_WORDS = getenv("FILTER_BADLISTED_WORDS", 0)

diff --git a/response_selectors/convers_evaluation_based_selector/Dockerfile b/response_selectors/convers_evaluation_based_selector/Dockerfile
@@ -52,4 +52,7 @@ ENV LANGUAGE ${LANGUAGE}
 COPY ./response_selectors/convers_evaluation_based_selector/ ./
 COPY ./common/ ./common/
 
+# wait for a server answer ( INTERVAL + TIMEOUT ) * RETRIES seconds after that change stutus to unhealthy
+HEALTHCHECK --interval=5s --timeout=5s --retries=3 CMD curl --fail 127.0.0.1:${SERVICE_PORT}/healthcheck || exit 1
+
 CMD gunicorn --workers=2 server:app
diff --git a/response_selectors/convers_evaluation_based_selector/requirements.txt b/response_selectors/convers_evaluation_based_selector/requirements.txt
@@ -7,3 +7,5 @@ sentry-sdk==0.12.3
 nltk==3.2.5
 jinja2<=3.0.3
 Werkzeug<=2.0.3
+six==1.16.0
+healthcheck==1.3.3
diff --git a/response_selectors/convers_evaluation_based_selector/server.py b/response_selectors/convers_evaluation_based_selector/server.py
@@ -10,6 +10,7 @@
 import numpy as np
 import sentry_sdk
 from flask import Flask, request, jsonify
+from healthcheck import HealthCheck
 from nltk.tokenize import sent_tokenize
 
 from common.greeting import greeting_spec, HI_THIS_IS_DREAM
@@ -41,6 +42,7 @@
 logger = logging.getLogger(__name__)
 
 app = Flask(__name__)
+health = HealthCheck(app, "/healthcheck")
 
 CALL_BY_NAME_PROBABILITY = float(getenv("CALL_BY_NAME_PROBABILITY", 0.5))  # if name is already known
 TAG_BASED_SELECTION = getenv("TAG_BASED_SELECTION", False)

diff --git a/response_selectors/llm_based_response_selector/Dockerfile b/response_selectors/llm_based_response_selector/Dockerfile
@@ -26,5 +26,7 @@ COPY ./response_selectors/llm_based_response_selector/ /src/
 WORKDIR /src
 COPY ./common/ ./common/
 
+# wait for a server answer ( INTERVAL + TIMEOUT ) * RETRIES seconds after that change stutus to unhealthy
+HEALTHCHECK --interval=5s --timeout=5s --retries=3 CMD curl --fail 127.0.0.1:${SERVICE_PORT}/healthcheck || exit 1
 
 CMD gunicorn --workers=1 server:app -b 0.0.0.0:${SERVICE_PORT} --timeout=1200
diff --git a/response_selectors/llm_based_response_selector/requirements.txt b/response_selectors/llm_based_response_selector/requirements.txt
@@ -6,3 +6,5 @@ numpy==1.25.0
 sentry-sdk==0.12.3
 jinja2<=3.0.3
 Werkzeug<=2.0.3
+six==1.16.0
+healthcheck==1.3.3
diff --git a/response_selectors/llm_based_response_selector/server.py b/response_selectors/llm_based_response_selector/server.py
@@ -11,6 +11,8 @@
 
 import sentry_sdk
 from flask import Flask, request, jsonify
+from healthcheck import HealthCheck
+
 from common.containers import get_envvars_for_llm, is_container_running
 from common.prompts import send_request_to_prompted_generative_service, compose_sending_variables
 from common.utils import is_toxic_or_badlisted_utterance
@@ -21,6 +23,7 @@
 logger = logging.getLogger(__name__)
 
 app = Flask(__name__)
+health = HealthCheck(app, "/healthcheck")
 
 GENERATIVE_SERVICE_URL = getenv("GENERATIVE_SERVICE_URL")
 while True:

diff --git a/response_selectors/ranking_and_intent_based_response_selector/Dockerfile b/response_selectors/ranking_and_intent_based_response_selector/Dockerfile
@@ -22,5 +22,7 @@ COPY ./response_selectors/ranking_and_intent_based_response_selector/ /src/
 WORKDIR /src
 COPY ./common/ ./common/
 
+# wait for a server answer ( INTERVAL + TIMEOUT ) * RETRIES seconds after that change stutus to unhealthy
+HEALTHCHECK --interval=5s --timeout=5s --retries=3 CMD curl --fail 127.0.0.1:${SERVICE_PORT}/healthcheck || exit 1
 
 CMD gunicorn --workers=1 server:app -b 0.0.0.0:${SERVICE_PORT} --timeout=1200
diff --git a/response_selectors/ranking_and_intent_based_response_selector/requirements.txt b/response_selectors/ranking_and_intent_based_response_selector/requirements.txt
@@ -6,3 +6,5 @@ numpy==1.25.0
 sentry-sdk==0.12.3
 jinja2<=3.0.3
 Werkzeug<=2.0.3
+six==1.16.0
+healthcheck==1.3.3
diff --git a/response_selectors/ranking_and_intent_based_response_selector/server.py b/response_selectors/ranking_and_intent_based_response_selector/server.py
@@ -10,6 +10,7 @@
 
 import sentry_sdk
 from flask import Flask, request, jsonify
+from healthcheck import HealthCheck
 from common.universal_templates import (
     is_any_question_sentence_in_utterance,
     if_chat_about_particular_topic,
@@ -31,6 +32,7 @@
 logger = logging.getLogger(__name__)
 
 app = Flask(__name__)
+health = HealthCheck(app, "/healthcheck")
 
 SENTENCE_RANKER_ANNOTATION_NAME = getenv("SENTENCE_RANKER_ANNOTATION_NAME")
 SENTENCE_RANKER_SERVICE_URL = getenv("SENTENCE_RANKER_SERVICE_URL")

diff --git a/response_selectors/ranking_based_response_selector/Dockerfile b/response_selectors/ranking_based_response_selector/Dockerfile
@@ -22,5 +22,7 @@ COPY ./response_selectors/ranking_based_response_selector/ /src/
 WORKDIR /src
 COPY ./common/ ./common/
 
+# wait for a server answer ( INTERVAL + TIMEOUT ) * RETRIES seconds after that change stutus to unhealthy
+HEALTHCHECK --interval=5s --timeout=5s --retries=3 CMD curl --fail 127.0.0.1:${SERVICE_PORT}/healthcheck || exit 1
 
 CMD gunicorn --workers=1 server:app -b 0.0.0.0:${SERVICE_PORT} --timeout=1200
diff --git a/response_selectors/ranking_based_response_selector/requirements.txt b/response_selectors/ranking_based_response_selector/requirements.txt
@@ -6,3 +6,5 @@ numpy==1.25.0
 sentry-sdk==0.12.3
 jinja2<=3.0.3
 Werkzeug<=2.0.3
+six==1.16.0
+healthcheck==1.3.3