Create a script to update the index and lookup file used to serve predictions. (#352)

* This script will be the last step in a pipeline to continuously update
  the index for serving.

* The script updates the parameters of the search index server to point
  to the supplied index files. It then commits them and creates a PR
  to push those commits.

* Restructure the parameters for the search index server so that we can use
  ks param set to override the indexFile and lookupFile.

* We do this because we want to be able to push a new index by doing
  ks param set in a continuously running pipeline
* Remove default parameters from search-index-server

* Create a dockerfile suitable for running this script.

code_search/Makefile

@@ -72,6 +72,17 @@ build-ui-gcb:
 	gcloud builds submit --machine-type=n1-highcpu-32 --project=kubeflow-ci --config=./build/build.ui.json \
 		--timeout=3600 ./build
 
+build-index-updater-gcb:
+	mkdir  -p build
+	jsonnet ./docker/index_updater/build.jsonnet --ext-str gitVersion=$(GIT_VERSION) --ext-str tag=$(TAG) \
+	  > ./build/build.index_updater.json
+	cp -r ./docker ./build/
+	cp -r ./src ./build/
+	rm -rf ./build/src/code_search/dataflow/cli/test_data
+	rm -rf ./build/src/code_search/t2t/test_data
+	gcloud builds submit --machine-type=n1-highcpu-32 --project=kubeflow-ci --config=./build/build.index_updater.json \
+		--timeout=3600 ./build
+
 # Build but don't attach the latest tag. This allows manual testing/inspection of the image
 # first.
 push-cpu: build-cpu

code_search/docker/index_updater/Dockerfile

@@ -0,0 +1,9 @@
+FROM ubuntu:xenial
+
+RUN apt-get update && apt-get install -y wget &&\
+    rm -rf /var/lib/apt/lists/*
+
+RUN wget -O /tmp/hub-linux-amd64-2.6.0.tgz https://github.com/github/hub/releases/download/v2.6.0/hub-linux-amd64-2.6.0.tgz && \
+	cd /usr/local && \
+	tar -xvf /tmp/hub-linux-amd64-2.6.0.tgz && \
+	ln -sf /usr/local/hub-linux-amd64-2.6.0/bin/hub /usr/local/bin/hub

code_search/docker/index_updater/README.md

@@ -0,0 +1,3 @@
+# Index Updater
+
+A Docker image and script suitable for updating the index served.

code_search/docker/index_updater/build.jsonnet

@@ -0,0 +1,26 @@
+// TODO(jlewi): We should tag the image latest and then
+// use latest as a cache so that rebuilds are fast
+// https://cloud.google.com/cloud-build/docs/speeding-up-builds#using_a_cached_docker_image
+{
+	
+	"steps": [      
+    {
+      "id": "build",
+      "name": "gcr.io/cloud-builders/docker",
+      "args": ["build", "-t", "gcr.io/kubeflow-examples/code-search/index_updater:" + std.extVar("tag"), 
+             	 "--label=git-versions=" + std.extVar("gitVersion"), 
+               "--file=docker/index_updater/Dockerfile", 
+               "."],
+    },
+    {
+      "id": "tag",
+      "name": "gcr.io/cloud-builders/docker",
+      "args": ["tag", "gcr.io/kubeflow-examples/code-search/index_updater:" + std.extVar("tag"), 
+               "gcr.io/kubeflow-examples/code-search/index_updater:latest",],
+      "waitFor": ["build"],
+    },    
+  ],
+  "images": ["gcr.io/kubeflow-examples/code-search/index_updater:" + std.extVar("tag"), 
+             "gcr.io/kubeflow-examples/code-search/index_updater:latest", 
+            ],
+}

code_search/docker/index_updater/update_index.sh

@@ -0,0 +1,95 @@
+#!/bin/bash
+#
+# This script creates a PR updating the nmslib index used by search-index-server.
+# It uses ks CLI to update the parameters.
+# After creating and pushing a commit it uses the hub github CLI to create a PR.
+#
+# The argument --base can be used to change the owner/org of the repo the PR is opened on.
+# To use the main kubeflow/examples repo use 
+# --base=kubeflow:master
+#
+# To use user alex's fork use
+# --base=alex/master
+set -ex
+
+DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" > /dev/null && pwd)"
+
+parseArgs() {
+  # Parse all command line options
+  while [[ $# -gt 0 ]]; do
+    # Parameters should be of the form
+    # --{name}=${value}
+    echo parsing "$1"
+    if [[ $1 =~ ^--(.*)=(.*)$ ]]; then
+    	name=${BASH_REMATCH[1]}
+    	value=${BASH_REMATCH[2]}
+
+    	eval ${name}="${value}"
+    elif [[ $1 =~ ^--(.*)$ ]]; then
+		name=${BASH_REMATCH[1]}
+		value=true
+		eval ${name}="${value}"
+    else
+    	echo "Argument $1 did not match the pattern --{name}={value} or --{name}"
+    fi
+    shift
+  done
+}
+
+usage() {
+	echo "Usage: update_index.sh --base=OWNER:branch --appDir=<ksonnet app dir> --env=<ksonnet environment> --indexFile=<index file> --lookupFile=<lookup file>"
+}
+
+parseArgs $*
+
+if [ ! -z ${help} ]; then
+	usage
+fi
+
+if [ -z ${dryrun} ]; then
+	dryrun=false
+fi
+
+# List of required parameters
+names=(appDir env lookupFile indexFile base)
+
+
+missingParam=false
+for i in ${names[@]}; do
+	if [ -z ${!i} ]; then
+		echo "--${i} not set"
+		missingParam=true		
+	fi	
+done
+
+if ${missingParam}; then
+	usage
+	exit 1
+fi
+cd ${appDir}
+ks param set --env=${env} search-index-server indexFile ${indexFile}
+ks param set --env=${env} search-index-server lookupFile ${lookupFile}
+git add .
+
+if (! ${dryrun}); then
+	git commit -m "Update the lookup and index file."
+	git push
+else
+	echo "dryrun; not committing to git."
+fi
+
+FILE=$(mktemp tmp.create_pull_request.XXXX)
+
+cat <<EOF >$FILE
+Update the lookup and index file.
+
+This PR is automatically generated by update_index.sh.
+
+This PR updates the index and lookup file used to serve
+predictions.
+EOF
+
+# Create a pull request
+if (! ${dryrun}); then
+hub pull-request --base=${base} -F ${FILE}
+fi

code_search/kubeflow/components/params.libsonnet

@@ -97,11 +97,9 @@
       indexFile: $.components["t2t-code-search"].workingDir + "/code_search_index.nmslib",
     },
     "search-index-server": {
+      // Most defaults should be defined in experiments.libsonnet.
+      // Parameters will be used to override those values.
       name: "search-index-server",
-      problem: $.components["t2t-code-search"].problem,
-      dataDir: $.components["t2t-code-search"].workingDir + "/data",
-      lookupFile: $.components["t2t-code-search"].workingDir + "/code_search_index.csv",
-      indexFile: $.components["t2t-code-search"].workingDir + "/code_search_index.nmslib",
       servingUrl: "http://t2t-code-search.kubeflow:8500/v1/models/t2t-code-search:predict",
       // 1 replica is convenient for debugging but we should bump after debugging.
       replicas: 1,

code_search/kubeflow/components/search-index-server.jsonnet

@@ -7,9 +7,10 @@ local experiments = import "experiments.libsonnet";
 
 local experimentName = baseParams.experiment;
 local experimentParams = experiments[experimentName];
-local params = baseParams + experimentParams + {
-  name: "search-index-server",
-};
+
+// baseParams override experiment parameters because we want to be able to set a new
+// index and csv file by doing ks param set.
+local params = experimentParams + baseParams;
 
 local deploymentSpec = {
   apiVersion: "extensions/v1beta1",

code_search/kubeflow/environments/cs_demo/params.libsonnet

@@ -1,14 +1,15 @@
-local params = std.extVar("__ksonnet/params");
-local globals = import "globals.libsonnet";
+local params = std.extVar('__ksonnet/params');
+local globals = import 'globals.libsonnet';
 local envParams = params {
   components+: {
-    "t2t-code-search"+: {
-    },
+    "t2t-code-search"+: {},
     "t2t-code-search-datagen"+: {
-      githubTable: "",
+      githubTable: '',
     },
     "submit-preprocess-job"+: {
-      githubTable: "",
+      githubTable: '',
+    },
+    "search-index-server"+: {
     },
   },
 };
@@ -18,4 +19,4 @@ local envParams = params {
     [x]: envParams.components[x] + globals
     for x in std.objectFields(envParams.components)
   },
-}
+}