Retail organizations constantly face financial pressure to increase sales while maintaining profit margins. Digital commerce creates new opportunities and a more competitive landscape for retailers by allowing them to reach a global customer base online, but it also exposes them to competition from larger online retailers. To be successful in this environment, retailers must not only have a strong online presence to keep up with their competition and gain market share, but also uplift the transactional customer experience to a more experiential one.

In today’s data-driven artificial intelligence-inspired business environment, organizations require complex IT infrastructure to support various functions such as prospecting, product development, marketing, and data analytics. Managing and scaling this infrastructure can be challenging, especially as needs evolve. As a result, many organizations are turning to Google Cloud as a solution for meeting their business goals, rather than simply expanding their in-house IT resources with more equipment and personnel. Specifically, retailers across the globe are betting on Kubernetes on Google Cloud to take advantage of secure, reliable, and scalable infrastructure.

Here are a few examples of worldwide retailers adapting to changing customer expectations using intelligent infrastructure solutions including Google Kubernetes Engine (GKE), the most scalable and automated fully managed Kubernetes from Google Cloud.

Haravan is a Vietnamese ecommerce platform that aims to improve the process of buying and selling products, allowing businesses to focus on creating and selling their products.

By using Google Cloud, Haravan helped small and medium-sized enterprises in Vietnam achieve double-digit growth, consistently met its 99.97% uptime commitment to clients, efficiently managed 5 times the normal amount of ecommerce activity, and facilitated the implementation of artificial intelligence-powered expansion plans.

“We have a guaranteed commitment to enable any volume of sales for clients over all channels, be it social networks, marketplaces, livestream, or website. Only Google Cloud, with the flexible autoscaling of GKE, gives us certainty to meet our guarantees even in the most massive Black Friday surges.” —Hung Le, VP of Software Engineering, Haravan

Loblaw is Canada’s food and pharmacy leader and the nation’s largest retailer. The company operates over 2,500 locations, including corporate, franchised, and associate-owned stores, and employs nearly 200,000 full- and part-time employees.

By using Google Cloud, boosted the performance of the online grocery platform, resulting in higher conversion rates and increased revenue, recovered up to 50% of Site Reliability Engineers’ time for innovation, introduced new, real-time personalization features and shopping conveniences for customers, and enhanced resiliency to protect customers and revenue.

“Moving our online grocery site to Google Cloud gave us a 4x performance increase and the capacity to handle up to three times the traffic; and we can scale up at any time.” —Hesham Fahmy, VP Technology, Loblaw

L.L.Bean is a North American retail company known for its boots and mail-order catalog, which dates back to 1912. The company has a strong online presence, with ecommerce accounting for $1 billion of its annual revenues of $1.6 billion. Like many other retailers, L.L.Bean is adopting an omnichannel sales strategy by interacting with customers through various channels including print, physical stores, its website, app, and social media.

By using Google Cloud, L.L.Bean enhanced customers’ online experience through faster page load times and access to transaction history, allowed for a focus on providing value to customers rather than managing infrastructure, and enabled the rapid release of cross-channel services by reducing development cycles.

“GKE has significantly streamlined the process of upgrading nodes and masters. By comparison, upgrading even minor releases of another container solution that L.L.Bean tested resulted in the need to rebuild that solution’s clusters four times.” —Randy Dyer, Enterprise Architect, L.L.Bean

LPP is a Polish fashion retailer established in 1991 by Lubianiec and Piechocki, whose initials make up the company’s name. LPP currently manages five clothing brands that are popular in 38 countries across Europe, Africa, and Asia, and has over 24,000 employees based in its main offices in Central and Eastern Europe. Growing demand for its ecommerce services led LPP to migrate from an on-premises setup to Google Cloud, harnessing automation to ensure great shopping experiences globally.

By using Google Cloud, LPP promoted a DevOps culture among developers through streamlined deployment of new features, provided 90% more time for engineers to work on innovative solutions instead of managing infrastructure, instantly created and updated new VMs, allowing developers to quickly launch new features, ensured a seamless online shopping experience by automatically adjusting capacity to meet demand.

“GKE enables us to deploy new features for our ecommerce sites very quickly. Previously, it took weeks to launch new instances for each brand. Today, it takes seconds: we simply launch a new machine, deploy the code, and changes are reflected automatically across our environment.” —Marek Maciejewski, Head of IT Service Operations, LPP

Noon.com, based in Riyadh, Saudi Arabia, is a local ecommerce marketplace focused on serving the Middle East. The company aims to become the top online retailer in the region, supporting the growth of a digital economy for both consumers and local businesses.

By using Google Cloud, Noon.com built its ecommerce platform to access self-managed services, allowed developers to establish a fully operational staging environment within two weeks, provided uninterrupted service to nearly four times as many daily users during busy seasons using autoscaling on GKE, used real-time data streaming on BigQuery to inform business decisions and personalize the customer experience, and achieved 99.999% availability with no downtime for planned maintenance or schema changes using a fully managed relational database.

“Google Cloud-managed services are playing a major role in enabling Noon.com customers to get their shopping done whenever they need it, without experiencing any delays or glitches, and without us having to lose sleep at night to ensure our platform is functioning as it should.” —Alex Nadalin, SVP of Engineering, Noon.com

In conclusion, the retail industry is constantly evolving and retailers must stay up-to-date with the latest technology and customer preferences to remain competitive. Digital commerce has changed the landscape of retail, allowing businesses to reach a global customer base but also increasing competition. The use of Kubernetes on Google Cloud can help retailers improve the customer experience, streamline internal processes, and make data-driven and AI-inspired decisions. By embracing these changes, retailers can stay ahead in a constantly evolving industry. Get started today with an exclusive workshop, Unlocking efficiency and innovation with Kubernetes on Google Cloud.

As we become increasingly reliant on technology to make decisions, geospatial data is becoming more critical than ever. It is a powerful resource that can be used to solve a variety of problems, from tracking the movement of goods, identifying interest areas and potential areas of disaster.

Geospatial data incorporates data with a geographic component, such as latitude and longitude coordinates, addresses, postal codes, or place names, and can be obtained from a variety of sources, including satellites, sensors, and surveys, making it an incredibly powerful tool with a broad range of applications.

One of the key features of BigQuery, Google Cloud’s serverless enterprise data warehouse, is its ability to analyze geospatial data. However, oftentimes geospatial data sits in a variety of public clouds, not just Google Cloud. To access it effectively, you need a multi-cloud analytics solution that lets you capitalize on the distinct capabilities of each cloud platform, while extracting insights and value from data sitting across multiple cloud platforms.

BigQuery Omni is a multi-cloud analytics solution that enables the analysis of data stored across public cloud environments, including Google Cloud, Amazon Web Services (AWS) and Microsoft Azure, without the need to transfer the data. With BigQuery Omni, users can employ the same SQL queries and tools used to analyze data in Google Cloud to analyze data in other clouds, making it easier to gain insights from all data, regardless of the storage location. For businesses using multiple clouds, BigQuery Omni is an excellent tool to unify analytics and optimize the value of data.

With BigQuery Omni, organizations can analyze location-based information or geographic components, such as latitude and longitude coordinates, addresses, postal codes, or place names without even copying their data to Google Cloud. For example, you could use BigQuery Omni to analyze data from a fleet of delivery vehicles to track their location and identify potential problems.

BigQuery Omni and geospatial data analysis

If you are working with geospatial data, BigQuery Omni is a powerful tool that can help you to get insights from your data. It is scalable, reliable, and secure, making it a great option for unifying your analytics and getting the most out of your data. 

Here are a few examples of ways organizations might use BigQuery Omni for geospatial data:

• A transportation company could use BigQuery Omni to analyze data from GPS sensors in its vehicles to track the movement of its fleet and identify potential problems.
• A retail company could use BigQuery Omni to analyze data from its point-of-sale systems to track customer behavior and identify trends.
• A government agency could use BigQuery Omni to analyze data from its weather sensors to track the movement of storms and identify areas at risk of flooding.

BigQuery Omni and geospatial data can be used together to gain insights into a variety of business problems. Specifically, some of the advantages of using BigQuery Omni and geospatial data include:

• Access to quality geospatial data: BigQuery supports loading of newline-delimited GeoJSON files and provides built-in support for loading and querying geospatial data. Data from public data sources like BigQuery public datasets, the Earth Engine catalog, and the United States Geological Survey (USGS) can be easily integrated into your BigQuery environment. Earth Engine has an integrated data catalog with a comprehensive collection of analysis-ready datasets, including satellite imagery and climate data. This data can be combined with proprietary data sources such as SAP, Oracle, Esri ArcGIS Server, Carto, and QGIS.
• Loading and preprocessing of geospatial data: BigQuery has built-in support for loading and querying geospatial data types, and you can use partner solutions such as FME Spatial ETL to load data.
• Working with different geospatial data types and formats: BigQuery supports a variety of file types and formats including WKT, WKB, CSV and GeoJSON.
• Coordinate reference systems: BigQuery’s geography data type is globally consistent. That means that your data is registered to the WGS84 reference system and your analyses can span a city block or multiple continents.

Overall, geospatial analytics with BigQuery Omni provides a wide range of technical capabilities for processing and analyzing geospatial data, making it a powerful tool for businesses that need to work with location-based data.

Analyzing geospatial data with BigQuery Omni

Imagine a retailer who has a large chain of department stores with locations all over the country. They are looking to expand their business and want to identify areas with high sales potential. They want a way to get a better understanding of their sales volume within specific geographic boundaries. To achieve this goal, the retailer turns to the GIS (Geographic Information System) functions built into BigQuery. Here are the steps what the retailer takes to analyze this dataset:

Step 1 : An initial orders dataset on AWS S3 contains 5.54 million rows, and with separate locations (300 rows) and zipcode (33144 rows) metadata files on AWS S3.

Orders Parquet files:

Location and Zipcode files:

Step  2 : The retailer uses BigQuery Omni to establish a connection between the data stored in AWS and BigQuery, enabling them to access the S3 datasets externally.

External Connection for AWS

External Table for orders

External Table for locations

External Table for zipcode

Step 3 : They combine the orders and locations datasets using BigQuery Omni, and remotely aggregate the data on AWS. Joining this dataset with geospatial datasets helps them derive geospatial coordinates.

The final aggregated dataset is reduced to 23 rows. Subsequently, they bring back the result dataset, which contains just 23 rows. This helps them reduce the extraction of millions of rows to just 23 rows for their geospatial analytics.

Select sales.store_city store_city,
sales.number_of_sales_last_10_mins number_of_sales_last_10_mins,
sales.store_zip store_zip,
 ST_GeogPoint(zip_lat_lng.longitude ,
   zip_lat_lng.latitude  ) geo
FROM (
 SELECT
   FORMAT_DATETIME("%X",
     CURRENT_DATETIME("America/Los_Angeles")) current_time,
   MAX(DATETIME(time_of_sale,
       "America/Los_Angeles")) time_of_last_sale,
   COUNT(1) number_of_sales_last_10_mins,
   locations.city store_city,
   locations.zip store_zip
 FROM
   `bqomni-blog.aws_locations.orders_small`  sales
 JOIN
   `bqomni-blog.aws_locations.locations` locations
 ON
   sales.store_id = locations.id
 GROUP BY
   locations.city,
   locations.zip ) sales
JOIN
   `bqomni-blog.aws_locations.zipcode` zip_lat_lng
ON
 cast(sales.store_zip as INT) = cast(zip_lat_lng.zipcode as INT)
 WHERE ST_WITHIN( ST_GeogPoint(zip_lat_lng.longitude , zip_lat_lng.latitude  ) ,ST_GeogFromText(zip_lat_lng.zipcode_geom ) )
 AND zip_lat_lng.state_name  = "New York"
 ORDER BY number_of_sales_last_10_mins

Aggregated Sales data by region

To build richer views of their sales volume data, the retailer uses BigQuery GeoViz integration, a powerful tool that allows for visualizing geographic data on maps.BigQuery Geo Viz is a web tool for visualization of geospatial data in BigQuery using Google Maps APIs. You can run a SQL query and display the results on an interactive map

BigQuery Geo view for sales data analysis

With the geo-tagged data in place, the retailer can now see regional sales volume, sales density, distribution by department by time, and distribution within department, all powered through BigQuery.

Satellite view

Benefits of using BigQuery Omni 

Beyond geospatial analysis, BigQuery Omni offers a number of benefits, including:

Reduced costs: BigQuery Omni’s ability to eliminate data transfers between clouds can help organizations reduce costs and simplify data management, making it a valuable tool for multi-cloud analytics. Also, the ability to access and analyze data across multiple clouds can reduce the need for data replication and synchronization, which can further simplify the ETL process and improve data consistency.

Unified governance: BigQuery Omni uses the same security controls as BigQuery, which include features such as encryption, access controls, and audit logs, to help protect data from unauthorized access.

Single pane for analytics: BigQuery Omni provides a single interface for querying data across all three clouds, which can simplify the process of analyzing data and reduce the need for organizations to use multiple analytics tools.

Flexibility: Analyze data stored in any of the supported cloud storage services, giving organizations the flexibility to work with the data they have regardless of where it’s located.

BigQuery Omni is a valuable tool for geospatial analysis because it allows you to analyze data from multiple sources without having to move the data. This can save you time and money, and it can also help you to get more accurate insights from your data.If you are looking for a way to improve the accuracy, efficiency, and decision-making of your business, using BigQuery Omni to analyze geospatial data can be a powerful tool.

References

• Omni-introduction 
• Omni-benefits 
• Omni-aws-create-connection
• Omni-azure-create-connection 
• BigQuery Omni Retail Demo on Looker
• Omni-pricing 
• Working with Geospatial Data
• Geospatial Analytics – Intro

Learn more about how BigQuery Omni can help your organization.

A few months ago, we wrote about how the first step to implementing Site Reliability Engineering (SRE) in an organization is getting leadership on board. So, let’s assume that you’ve gone ahead and done that. Now what? What are some concrete steps you can take to get the SRE ball rolling? In this blog post, we’ll take a look at what you as an IT leader can do to fast-track SRE within your team. 

Step 1: Start small and iterate 

“Rome wasn’t built in a day,” the saying goes, but you do need to start somewhere. When it comes to implementing SRE principles, the approach that I (and my team) found to be the most effective is to start with a proof of concept, learn from our mistakes, and iterate!

Start by identifying a relevant application and/or team 

There are many factors that go into choosing a specific team or application for your SRE proof of concept. Most of the time, though, this is a strategic decision for the organization, which is outside the scope of this article. Possible candidates can be a team shifting from traditional operations or DevOps to SRE, or a need to increase reliability to a business-critical product. No matter the reason, it’s crucial to select an application that is:

1. Critical to the business. Your customers should care deeply about its uptime and reliability. 
2. Currently in development. Pick an application in which the business is actively investing resources. 
3. In a perfect world, the application provides data and metrics regarding its behaviour. 

Conversely, stay away from proprietary software. If the application wasn’t built by you, it’s not a good candidate for SRE! You need the ability to make strategic decisions about—and engineering changes to—the application as needed. 

Pro tip: In general, if you have workloads both on-premises and in the cloud, try to start with the cloud-based app. If your engineers come from a traditional operations environment, changing their thinking away from ‘bare metal’ and infrastructure metrics will be easier for a cloud-based app, as managed infrastructure turns practitioners into users and forces them to consume it like developers (APIs, infrastructure as code, etc.)

Remember: Set realistic goals. Discouraging your team with unrealistic expectations early on will have a negative effect on the initiative. 

Step 2: Empower your teams

Implementing SRE principles requires fostering a learning culture, and in that regard, team enablement means both training them, i.e., in regards to knowledge, as well as empowering them.

Building a training program is a topic in and of itself, but it’s important to think about an enablement strategy at an early stage. Especially in large organizations, you need to address topics like internal upskilling, hiring and scaling the team as well as onboarding and creating a learning community. 

Your enablement strategy should also accommodate employees at different levels and in different functions. For example, higher leadership’s training will look very different from practitioners’ training. Leadership’s education should be sufficient to get buy-in and to be able to make organizational decisions. To drive change in the entire organization, additional training to leadership on cultural concepts and practices might be required.

LEARN MORETraining Site Reliability EngineersWhat your organization needs to create a learning program

When it comes to engineering leadership and/or middle management (managers that manage managers), training should be a combination of  high-level cultural concepts to help foster the required culture, and technical SRE practices that are deep enough to understand prioritization, resource allocation, process creation, and future needs.

When it comes to practitioners, ideally you want the entire organization to be aligned both from a knowledge perspective as well as culturally. But as we’ve mentioned earlier, it’s best to start simple, with just one team.

The starting point for those teams should be to understand reliability and key concepts like SLAs, SLOs, SLIs and error budgets. These are important because SRE is focused on the customer experience. Measuring whether systems meet customer expectations requires a shift in mindset and can take time.

LEARN MOREStart is with this Coursera courseThis course teaches the theory of Service Level Objectives (SLOs), a principled way of describing and measuring the desired reliability of a service.

After identifying your first application and/or the team responsible for it, it’s time to identify the app’s user journeys, the set of interactions a user has with a service to achieve a single goal—for example, a single click or a multi-step pipeline, and rank them according to business impact. The most critical ones are called Critical User Journeys (CUJ), and these are where you should start  drafting SLO/SLIs.

Pro tip: There are some general technical practices that can help you embrace SRE faster. For example, using less repos rather than more can help you reduce silos within the organization and better utilize resources. 

Likewise, prioritizing automatic processes and self-healing systems can benefit reliability, but also team satisfaction, helping the organization retain talent.

LEARN MORESRE Public Resources for GCP CustomersHow can you get started with SRE as an engineer/practitioner?

Final note: Similar to the way that you make architecture decisions, your chosen technology, solutions and implementation tools should enable you to do what you are trying to do and not vice versa. 

Step 3: Scale those learnings 

After you establish these SRE practices with one or a few teams, the next step is to think about building an SRE community and formalized processes across the organization. In some organizations, you can do this in parallel to the end of step 2, and in some organizations, only after you have a few successful implementations under your belt.

In this phase, you’ll probably want to address community, culture, enablement and processes. You will need to address them all, especially as they are intertwined, but which one you prioritize will depend on your organization.

Creating an SRE community in the organization is important both from a learning perspective, but also to establish a knowledge base of best practices, train subject-matter experts, help create needed guardrails, and align processes. 

Building a community goes hand in hand with fostering an empowered culture and training teams. The idea is that early adopters are ambassadors for SRE who share their learnings and train other teams in the organization. 

It is also useful to identify potential ambassadors or champions in individual development teams who are passionate about SRE and will help with the adoption of those practices.

It is also crucial to create repeatable trainings for each functional role, including onboarding sessions. Onboarding new team members is a critical aspect of training and fostering an empowered SRE culture. Therefore it is vital to be mindful about your onboarding process and make sure that the knowledge is not lost when team members change roles.

LEARN MOREDeploying SRE training best practices to productionHow to “SRE” an SRE training program

During this phase, you also want to foster an org-wide culture that promotes psychological safety, accepts failure as normal and enables the team to learn from mistakes. For that, leadership must model the desired culture and promote transparency. 

Finally, having structured and formalized processes can help reduce the stress around emergency response—especially being on-call. Processes can also provide clarity and make teams more collaborative and effective. 

In order to have the most impact, start by prioritizing the most painful areas under your team’s remit—for example, clean up noisy alerts to avoid (or address) alert fatigue, automate your change management processes and involve only the necessary people to save team bandwidth. Team members shouldn’t work on software engineering projects while doing on-call incident management, and vice-versa. Make sure they have enough bandwidth to do both, separately.  Similar to other areas, you want to use data to drive your decisions.  As such, identify where your teams spend the most time, and for how long. 

If you find that it is challenging to collect this kind of data, be it quantitative or qualitative, a good starting point is often your emergency response processes, as those have a direct impact on the business, especially around the escalation process, incident management and related policies. 

Pro tip: All the above practices contribute to reducing silos and align goals across the organization; those should include also your vendors and engineering partners. To that end, make sure your contracts with them capture those goals as well.

Step 4: Embody a data-driven mindset

Starting the SRE journey can take time, even if you’re just implementing it for one team. Two quick wins that you can start with that will make a positive impact are collecting data and doing blameless postmortems.

In SRE we try to be as data-driven as possible, so creating a measurement culture in your organization is crucial. When prioritizing data collection, ideally look for data that represents the customer experience. Collecting that data will help you identify your gaps and help you prioritize according to business needs and by extension your customer expectations.

LEARN MORELogging, Monitoring and Observability in Google CloudLearn how to monitor, troubleshoot, and improve your infrastructure and application performance.

Another thing that you can do is run or improve postmortems, which are an essential way of learning from failure and fostering a strong SRE culture. From our experience, even organizations that do run postmortems can benefit from them much more with a few minor improvements. It is important to remember that postmortems should be blameless in order to make the team feel safe to share and learn from failures. And to make tomorrow better than today, i.e., not repeat the same problems, it’s important that postmortems include action items and are assigned to an owner. 

Creating a shared repository for postmortems can have a tremendous impact on the team: it increases transparency, reduces silos, and contributes to the learning culture. It also shows the team that the organization “practices what it preaches.” Implementing a repository can be as easy as creating a shared drive.

Pro tip: Postmortems should be blameless and actionable.

LEARN MOREImplementing SRE practices using Cloud OperationsHow can you get started with SRE as an engineer/practitioner?

On the SRE fast track

Of course, no two organizations are alike, and no two SRE teams are either. But by following these steps, you can help get your team on the path to SRE success faster. To learn more about developing an effective SRE practice, check out the following resources. 

• Collection of SRE Public resources
• Google Professional Services SRE packages

This post shows how to build an image search utility using natural language queries. Our aim is to use different GCP services to demonstrate this. At the core of our project is OpenAI’s CLIP model. It makes use of two encoders – one for images and one for texts. Each encoder is trained to learn representations such that similar images and text embeddings are projected as close as possible.

We will first create a Flask-based REST API capable of handling natural language queries and matching them against relevant images. We will then demonstrate the use of the API through a Flutter-based web and mobile application. Figure 1 shows how our final application would look like:

All the code shown in this post is available as a GitHub repository. Let’s dive in. 

Application at a high-level

Our application will take two queries from the user:

• Tag or keyword query. This is needed in order to pull a set of images of interest from Pixabay. You can use any other image repositories for this purpose. But we found Pixabay’s API to be easier to work with. We will cache these images to optimize the user experience. Suppose we wanted to find images that are similar to this query: “horses amidst flowers”. For this, we’d first pull in a few “horse” images and then run another utility to find out the images that best match our query.  
• Longer or semantic query that we will use to retrieve the images from the pool created in the step above. These images should be semantically similar to this query. 

Note: Instead of two queries, we could have only taken a single long query and run named-entity extraction to determine the most likely important keywords to run the initial search with. For this post, we won’t be using this approach. 

Figure 2 below depicts the architecture design of our application and the technical stack used for each of the components.

Figure 2 also presents the core logic of the API we will develop in bits and pieces in this post. We will deploy this API on a Kubernetes cluster using the Google Kubernetes Engine (GKE). The following presents a brief directory structure of our application code-base:

Next, we will walk through the code and other related components for building our image search API. For various machine learning-related utilities, we will be using PyTorch. 

Building the backend API with Flask

First, we’d need to fetch a set of images with respect to user-provided tags/keywords before performing the natural language image search. The utility below from the pixabay_utils.py script can do this for us:

  def fetch_images_tag(pixabay_search_keyword, num_images):
    """
    Fetches images from Pixabay w.r.t a keyword.
    :param pixabay_search_keyword: Keyword to perform the search on Pixabay.
    :param num_images: Number of images to retrieve.
    :return: List of PIL images.
    :return: List of image URLs.
    """
    query = (
        PIXABAY_API
        + "&q="
        + pixabay_search_keyword.lower()
        + "&image_type=photo&safesearch=true&per_page="
        + str(num_images)
    )
    
    response = requests.get(query)
    output = response.json()

    all_images = []
    all_image_urls = []

    for each in output["hits"]:
        imageurl = each["webformatURL"]
        response = requests.get(imageurl)
        image = Image.open(BytesIO(response.content)).convert("RGB")
        all_images.append(image)
        all_image_urls.append(imageurl)

    return (all_images, all_image_urls)

Note that all the API utilities are logging relevant information. But for brevity, we have omitted the lines of code responsible for that. Next, we will see how to invoke the CLIP model and select the images that would best match a given query semantically.  For this, we’ll be using Hugging Face, an easy-to-use Python library offering state-of-the-art NLP capabilities. We’ll collate all the logic related to this search inside a SimilarityUtil class:

  class SimilarityUtil:
    def __init__(self):
        self.model = CLIPModel.from_pretrained(CLIP_MODEL)
        self.processor = CLIPProcessor.from_pretrained(CLIP_PREPROCESSOR)
        self.device = "cuda" if torch.cuda.is_available() else "cpu"

    def perform_sim_search(self, images, query_phrase, top_k=3):
        """
        Performs similarity search between the images and query.
        :param images: A list of PIL images initially retrieved with
        respect to some entity e.g. Tiger.
        :param query_phrase: A list containing a single text query,
        e.g. "Tiger drinking water".
        :param top_k: Number of top images to return from `images`.
        :return: Top-k indices matching the query semantically and
        their similarity scores.
        """
        model = self.model.to(self.device)
        # Obtain the text-image similarity scores
        with torch.no_grad():
            inputs = self.processor(
                text=[query_phrase], images=images, return_tensors="pt", padding=True
            )
            inputs = inputs.to(self.device)
            outputs = model(**inputs)

        # Image-text similarity scores
        logits_per_image = outputs.logits_per_image.cpu()
        (top_indices, top_scores) = self.sort_scores(logits_per_image, top_k)

        return (top_indices, top_scores)

    def sort_scores(self, scores, top_k):
        """
        Sorts the scores in a descending manner.
        :param scores: Scores to sort through.
        :param top_k: Number of top scores to return.
        :return: Top-k scores and their indices.
        """
        values, indices = scores.squeeze().topk(top_k)
        top_indices, top_scores = [], []

        for score, index in zip(values, indices):
            top_indices.append(int(index.numpy()))
            score = score.numpy().tolist()
            top_scores.append(round(score, 3))

        return (top_indices, top_scores)

CLIP_MODEL uses a ViT-base model to encode the images for generating meaningful embeddings with respect to the provided query. The text-based query is also encoded using A Transformers-based model for generating the embeddings. These two embeddings are matched with one another during inference. To know more about the particular methods we are using for the CLIP model please refer to this documentation from Hugging Face. 

In the code above, we are first invoking the CLIP model with images and the natural language query. This gives us a vector (logits_per_image) that contains the similarity scores between each of the images and the query. We then sort the vector in a descending manner. Note that we are initializing the CLIP model while instantiating the SimilarityUtil to save us the model loading time. This is the meat of our application and we have tackled it already. If you want to interact with this utility in a live manner you can check out this Colab Notebook. 

Now, we need to collate our utilities for fetching images from Pixabay and for performing the natural language image search inside a single script – perform_search.py. Following is the main class of that script:

  class Searcher:
    def __init__(self):
        self.similarity_model = SimilarityUtil()

    def get_similar_images(self, keyword, semantic_query, pixabay_max, top_k):
        """
        Finds semantically similar images.
        :param keyword: Keyword to search with on Pixabay.
        :param semantic_query: Query to find semantically similar images retrieved from Pixabay.
        :param pixabay_max: Number of maximum images to retrieve from Pixabay.
        :param top_k: Top-k images to return.
        :return: Tuple of top_k URLs and the similarity scores of the images present inside the URLs.
        """
        images_redis_key = keyword + "_images"
        urls_redis_key = keyword + "_urls"

        if redis_client.exists(images_redis_key) and redis_client.exists(
            urls_redis_key
        ):
            keyword_images = redis_client.get(images_redis_key)
            keyword_image_urls = redis_client.get(urls_redis_key)
        else:
            (keyword_images, keyword_image_urls) = fetch_images_tag(
                keyword, pixabay_max
            )
            redis_client.set(images_redis_key, keyword_images)
            redis_client.set(urls_redis_key, keyword_image_urls)

        (top_indices, top_scores) = self.similarity_model.perform_sim_search(
            keyword_images, semantic_query, top_k
        )

        top_urls = [keyword_image_urls[index] for index in top_indices]

        return (top_urls, top_scores)

Here, we are just calling the utilities we had previously developed to return the URLs of the most similar images and their scores. What is even more important here is the caching capability. For that, we combined GCP’s MemoryStore and a Python library called direct-redis. More on setting up MemoryStore later. 

MemoryStore provides a fully managed and low-cost platform for hosting Redis instances. Redis databases are in memory and light-weight making them an ideal candidate for caching. In the code above, we are caching the images fetched from Pixabay and their URLs. So, in the event of a cache hit, we won’t need to call the CLIP model and this will tremendously improve the response time of our API. 

Other options for caching

We can cache other elements of our application. For example, the natural language query. When searching through the cached entries to determine if it’s a cache hit, we can compare two queries for semantic similarity and return results accordingly. 

Consider that a user had entered the following natural language query: “mountains with dark skies”. After performing the search, we’d cache the embeddings of this query. Now, consider that another user entered another query: “mountains with gloomy ambiance”. We’d compute its embeddings and run a similarity search with the cached embeddings. We’d then compare the similarity scores with respect to a threshold and parse the most similar queries and their corresponding results. In case of a cache miss, we’d just call the image search utilities we developed above. 

When working on real-time applications we often need to consider these different aspects and decide what enhances the user experience and maximizes business at the same time. 

All that’s left now for the backend is our Flask application – main.py:

  @app.route("/search", methods=["GET"])
def get_images():
    tag = request.args.get("t").lower()
    query = request.args.get("s_query").lower()
    top_k = request.args.get("k")

    (top_urls, top_scores) = searcher.get_similar_images(
        tag, query, MAX_PIXABAY_SEARCH, int(top_k)
    )

    return jsonify({"top_urls": top_urls, "top_scores": top_scores})

Here we are first parsing the query parameters from the request payload of our search API.  We are then just calling the appropriate function from perform_search.py to handle the request. This Flask application is also capable of handling CORS. We do this via the flask_cors library:

  cors = CORS(
    app,
    resources={
        r"/search/*": {"origin": "*"},
        r"/test/*": {"origin": "*"},
    },
)

And this is it! Our API is now ready for deployment. 

Deployment with Compute Engine and GKE

The reason why we wanted to deploy our API on Kubernetes is because of the flexibility Kubernetes offers for managing deployments. When operating at scale, auto scalability and load balancing are very important. With the comes the requirement of security — we’d not want to expose the utilities for interacting with any internal services such as databases. With Kubernetes, we can achieve all these easily and efficiently. 

GKE provides secured and fully managed functionalities for operationalizing Kubernetes clusters. Here are the steps to deploy the API on GKE at a glance:

• We first build a Docker image for our API and then push it to the Google Container Registry (GCR).
• We then create a Kubernetes cluster on GKE and initialize a deployment.
• We then add scalability options.
• If any public exposure is needed for the API, we then tackle it. 

We can assimilate all the above into a shell script – k8s_deploy.sh:

  ## Docker build and push ## 
# We are inside the `server` directory
docker build -t gcr.io/${PROJECT_ID}/search_service .
docker push gcr.io/${PROJECT_ID}/search_service

## Deploy on a GKE cluster ##
# Configure the Docker command-line tool to authenticate to Container Registry
gcloud auth configure-docker

# Create a Kubernetes cluster
gcloud container clusters create image-search-nlp 

# Create a deployment
kubectl create deployment clip-search --image=gcr.io/${PROJECT_ID}/search_service

# Number of worker replicas
kubectl scale deployment clip-search --replicas=3

# HorizontalPodAutoscaler resource
kubectl autoscale deployment clip-search --cpu-percent=80 --min=1 --max=5

# Expose deployment
kubectl  expose deployment clip-search --name=clip-search-service --type=LoadBalancer --port 80 --target-port 8080

These steps are well explained in this tutorial that you might want to refer to for more details. We can configure all the dependencies on our local machine and execute the shell script above. We can also use the GCP Console to execute it since a terminal on the GCP Console is pre-configured with the system-level dependencies we’d need. In reality, the Kubernetes cluster should only be created once and different deployment versions should be created under it. 

After the above shell script is run successfully, we can run kubectl get service to know the external IP address of the service we just deployed:

  NAME                       CLUSTER-IP    EXTERNAL-IP     PORT(S)          AGE
clip-search-service      10.3.251.122    203.0.113.0     80:30877/TCP     10s

We can now consume this API with the following base URI: http://203.0.113.0/. If we wanted to deal with only http-based API requests, then we are done here. But secured communication is often a requirement in order for applications to operate reliably. In the following section, we are to discuss how to configure the additional items to allow our Kubernetes cluster to allow https requests.  

Configurations for handling https requests with GKE

A secure connection is almost often a  must-have requirement in modern client/server applications. The front-end Flutter application would be hosted on GitHub Pages for this project, and it requires https-based connection as well. Even if configuring https connection particularly for a GKE-based cluster can be considered a chore, its setup might seem daunting at first.

There are six steps to configure https connection in the GKE environment: 

1. You need to have a domain name, and there are a lot of inexpensive options that you can buy. For instance, mlgde.com domain for this project is acquired via Gabia which is a Korean service provider.
2. A reserved (static) external IP address has to be acquired via gcloud command or GCP console. 
3. You need to bind the domain name with the acquired external IP address. This is a platform-specific configuration that issued the domain name to you. 
4. There is a special ManagedCertificate resource which is specific to the GKE environment. ManagedCertificate resource specifies the domain that the SSL certificate will be created for, so you need this. 
5. An Ingress resource should be created by listing the static external IP address, ManagedCertificate resource, and the service name and port which the incoming traffic will be routed to. The Service resource could remain the same as in the above section with only changes from LoadBalancer to ClusterIP. 
6. Last but not least, you need to modify the existing Flask application and Deployment resource to support liveness and readiness probes which are used to check the health status of the Deployment. The Flask application side can be simply modified with the flask-healthz Python package, and you only need to add livenessProbe and readinessProbe sections in the Deployment resource. In the code example below, the livenessProbe and readinessProbe are checked via /alive and /ready endpoints respectively.

  from flask_healthz import healthz
from flask_healthz import HealthError


app = Flask(__name__)
app.register_blueprint(healthz, url_prefix="/")


def printok():
   print("Everything is fine")


def liveness():
   try:
       printok()
   except Exception:
       raise HealthError("Can't connect to the file")


def readiness():
   try:
       printok()
   except Exception:
       raise HealthError("Can't connect to the file")


app.config.update(
   HEALTHZ = {
       "alive": "main.liveness",
       "ready": "main.readiness",
   }
)

One thing to be careful of is the initialDelaySeconds attribute of the probes. It is uncommon to configure this attribute with a big number, but it could be bigger than 90 – 120 seconds depending on the size of the model to be used. For this project, it is configured in 90 seconds in order to wait until the CLIP model is fully loaded into memory (full YAML script here).

  livenessProbe:
 httpGet:
   path: /alive
   port: 8080
 initialDelaySeconds: 90
 periodSeconds: 10

readinessProbe:
 httpGet:
   path: /ready
   port: 8080
 initialDelaySeconds: 90
 periodSeconds: 10

Again, these steps may seem daunting at first, but it will become clear when you have done it once. Here is the official document for Using Google-managed SSL certificates You can find all the GKE-related resources used in this project here. 

Once every step is completed you should be able to see your server application running on the GKE environment. Please make sure to run kubectl apply command whenever you create Kubernetes resources such as Deployment, Service, Ingress, and ManagedCertificate, and it is important to wait for more than 10 minutes until the ManagedCertifcate provisioning is done. 

You can run gcloud compute addresses list command to find out the static external IP address that you have configured.

  NAME  ADDRESS/RANGE  TYPE      PURPOSE  NETWORK  REGION  SUBNET  STATUS
gde      34.149.231.34       EXTERNAL                                                                    IN_USE

Then, the IP address has to be mapped to the domain. Figure 3 is a screenshot of a dashboard from where we got the mlgde.com domain. It clearly shows mlgde.com is mapped to the static external IP address configured in GCP.

In case you’re wondering why we didn’t deploy this application on App Engine, well that is because of the compute needed to execute the CLIP model. App Engine instance won’t fit in that regime. We could have also incorporated compute-heavy capabilities via a VPC Connector. That is a design choice that you and your team would need to consider. In our experiments, we found the GKE deployment to be easier and suitable for our needs. 

Infrastructure for the CLIP model

As mentioned earlier, at the core of our application is the CLIP model. It is computationally a bit more expensive than the regular deep learning models. This is why it makes sense to have the hardware infrastructure set up accordingly to execute it. We ran a small benchmark in order to see how a GPU-based environment could be beneficial here. 

We ran the CLIP on a Tesla P100-based machine and also on a standard CPU-only machine 1000 times. The code snippet below is the meat of what we executed:

  DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

start_time = time.time()
for _ in range(1000):
    with torch.no_grad():
        model = model.to(DEVICE)
        inputs = processor(text=[semantic_search_phrase], 
                        images=all_images, return_tensors="pt", padding=True)
        inputs = inputs.to(DEVICE)
        outputs = model(**inputs)
end_time = time.time() - start_time
print(f"Total time: {end_time:.3f} seconds.")

As somewhat expected, with the GPU, the code took 13 minutes to complete execution. With no GPU, it took about 157 minutes.

It is uncommon to leverage GPUs for model prediction because of cost restrictions, but sometimes we have to access GPUs for deploying a big model like CLIP. We configured a GPU-based cluster on GKE and compared the performance differences with and without it. It took about 1 second to handle a request with GPU and MemoryStore cache while it took more than 4 seconds with MemoryStore only (without the GPUs). 

For the purposes of this post, we used a CPU-based cluster on Kubernetes. But It is easy to configure GPU usage in a GKE cluster. This document shows you how to do so. For a short summary, there are two steps. First, a node should be configured with GPUs when creating a GKE cluster. Second, GPU drivers should be installed in GKE nodes. You don’t need to visit and manually install GPU drivers for each node by yourself. Rather you can simply apply the DaemonSet resource to GKE as described here.

Setting up MemoryStore

In this project, we first query the general concept of images to Pixabay, then we filter the images with a semantic query using CLIP. It means we can cache the initially retrieved images from Pixabay for the next specific semantic query. For instance, you may want to search with “gentleman wearing tie” at first, then you may want to retry searching for “gentleman wearing glass”. In this case, the base images remain all the same, so they could be stored in a cache server like Redis. 

MemoryStore is a GCP service wrapping the Redis which is an in-memory data store, so you can simply use a standard Redis Python package for accessing it. The only thing to be careful about when provisioning a MemoryStore Redis instance is to make sure it is in the same region where your GKE cluster or Compute Engine instance is.

The code snippet below shows how to make a connection to the Redis instance in Python. Nothing specific to GCP, but you only need to be aware of the usage of the standard redis-py package. 

  # REDISHOST is the IP address to the MemoryStore instance
redis_host = os.environ.get("REDISHOST", "localhost")
redis_port = int(os.environ.get("REDISPORT", 6379))
redis_client = redis.StrictRedis(host=redis_host, port=redis_port)

After creating a connection, you can store and retrieve data from MemoryStore. There are more advanced use cases of Redis, but we only used exists, get, and set methods for the demonstration purpose. These methods should be very familiar if you know maps, dictionaries, or other similar data structures. For the code portion that uses Redis-related utilities, please refer to the Searcher Python class we discussed in an earlier section. 

In the URLs below, you can find side-by-side comparisons of using MemoryStore:

• Without MemoryStore: https://youtu.be/7B88Eyrd-4s
• With MemoryStore (1st try): https://youtu.be/LE6xeEIRuMM
• With MemoryStore (2nd try): https://youtu.be/rRfK17sdk84 

Putting everything together

All that’s left now is to collate the different components we developed in the sections above and deploy our application with a frontend. All the frontend-related code is present here. 

• The front-end application is written in the Flutter development kit. The main screen contains two text fields for queries to Pixabay and CLIP model respectively. When you click the “Send Query” button, it will send out a RestAPI request to the server. After receiving the result back from the server, the retrieved images from the semantic query will be displayed at the bottom section of the screen. 
  • Please note that a Flutter application can be deployed to various environments including desktop, web, iOS, and Android. In order to keep as simple as possible, we chose to deploy the application to the GitHub Pages. Whenever there is any change to a client-side source directory, the GitHub Action will be triggered to build a web page and deploy the latest version to the GitHub Pages. 

Our final application is deployed here and it looks like so:

Note that due to constraints, the above-mentioned URL will only be live for one or two months.  

It is also possible to redeploy the back-end application with a GitHub Action. 

• The very first step is to craft a Dockerfile like below. Since Python is a scripting language, and there are lots of heavy packages that the application is dependent on, it is important to cache the steps. For instance, installing the dependencies should be separated from other commands.

  FROM pytorch/pytorch:latest
WORKDIR /app

# install the dependencies
COPY requirements.txt requirements.txt
RUN pip install -r requirements.txt

COPY . .  

# set up environment variables
ENV PIXABAY_API_KEY="..."
ENV REDIS_IN_USE="true"
ENV REDISHOST="..."

# expose port that Flask app is listening on
EXPOSE 8080

# run the Flask app
CMD [ "python3", "main.py" ]

• With the Dockerfile defined, we can use a GitHub Action like this for automatic deployment. 

Edge cases

Since the CLIP model is pre-trained on a large corpus of image and text pairs it’s likely that it may not generalize well to every natural language query we throw at it. Also, because we are limiting the number of images on which the CLIP model can operate, this somehow restricts the expressivity of the model.

We may be able to improve the performance for the second situation by increasing the number of images to be pre-fetched and by indexing them into a low-cost and high-performance database like Datastore. 

Costs

In this section, we wanted to provide the readers a breakdown of the costs they might incur in order to consume the various services used throughout the application. 

• Frontend hosting
  • The front-end application is hosted on GitHub Pages, so there is no expenditure for this.
• Compute Engine
  • With an e2-standard-2 instance type without GPUs, the cost is around $48.92 per month. In case you want to add a GPU (NVIDIA K80), the cost goes up to $229.95 per month.
• MemoryStore
  • The cost for MemoryStore depends on the size. With 1GB of space, the cost is around $35.77 per month, and whenever you add more GBs the cost will be doubled.
• Google Kubernetes Engine
  • The monthly cost for a 3 node GKE cluster with n2-standard-2 (vCPUs: 2, RAM: 8GB without GPUs) is about $170.19. If you add one GPU (NVIDIA K80) to the cluster, the cost goes up to $835.48.

While you may think that is a lot cost-wise, it is good to know that Google gives away free $300 credits when you create a new GCP account. It is still not enough for leveraging GPUs, but it is enough to learn and experiment with GKE and MemoryStore usage.

Conclusion

In this post, we walked through the components needed to build a basic image search utility for natural language queries. We discussed how these different components are connected to each other. Our image search API is able to utilize caching and was deployed on a Kubernetes cluster using GKE. These elements are essential when building a similar service to cater to a much bigger workload. We hope this post will serve as a good starting point for that purpose. Below are some references on similar areas of work that you can explore:

• Building a real-time embeddings similarity matching system
• Detecting image similarity using Spark, LSH and TensorFlow

Acknowledgments: We are grateful to the Google Developers Experts program for supporting us with GCP credits. Thanks to Karl Weinmeister and Soonson Kwon of Google for reviewing the initial draft of this post. 

Today, many applications in organizations’ data centers run on Windows Server. Modernizing these traditional Windows apps onto Kubernetes promises a host of benefits: a consistent platform across environments, better portability, scalability, availability, simplified management and speed of deployment, just to name a few. But how? Rewriting traditional .NET applications to run on Linux with .NET Core can be challenging and time-consuming. There is, however, a lower-toil, more developer friendly option.

Last year, we announced support for Windows Server containers running on Google Kubernetes Engine (GKE), our cloud-based managed Kubernetes service, which lets you take the advantage of containers without porting your apps to .NET core or rewriting them for Linux. Today, we’re going a step further with support for Windows Server containers on Anthos clusters on VMware in your on-premises environment. Now available in preview, you can consolidate all your Windows operations across on-prem and Google Cloud.

Bringing Windows Server support to our family of Kubernetes-based services—GKE running on Google Cloud, and Anthos everywhere—with the same experience, lets you modernize apps faster and achieve a consistent development and deployment experience across hybrid and cloud environments. Further, by running Windows and Linux workloads side by side, you get operational consistency and efficiency—no need to have multiple teams specializing in different tooling or platforms to manage different workloads. The single-pane-of-glass view and the ability to manage policies from a central control plane simplifies the management experience, while bin packing multiple Windows applications drives better resource utilization, leading to infrastructure and license savings.

With all these benefits, it’s no surprise that customers such as Thales, a French multinational firm specializing in aerospace and security services, have been able to reap significant benefits by moving Windows applications to GKE. 

“We moved our Windows applications from VMs to Windows containers on GKE and now have a unified mechanism for Linux and Windows-based application management, scaling, logging, and monitoring. Earlier, setting up these applications in VMs and configuring them for high availability used to take up to a week, and the applications were not easily scalable,” said Najam Siddiqui, Solutions Architect at Thales. “Now with GKE, the setup takes only a few minutes. GKE’s automatic scaling and built-in resiliency features make scaling and high-availability setup seamless. Also, manually maintaining the VMs and applying security patches used to be tedious, which is now handled by GKE.” 

Let’s take a deeper look at the architecture that lets you run your Windows container-based workloads on-prem. 

Windows Server running on-prem with Anthos 

The diagram below illustrates the high-level architecture of running Windows container-based workloads in an on-prem GKE cluster with Anthos. Windows server node-pools can be added to an existing or new Anthos cluster. Kubelet and Kube-proxy run natively on Windows nodes, allowing you to run mixed Windows and Linux containers in the same cluster. The admin cluster and the user cluster control plane continue to be Linux-based, providing you a consistent orchestration experience and management ease across Windows and Linux workloads.

Get started today

When considering modernizing your on-prem Windows estate, we recommend running Windows Server containers on Anthos in your own data center. If you are new to Anthos, the Anthos getting started page and the Coursera course on Architecting Hybrid Cloud with Anthos are good places to start. You can also find detailed documentation on our website, and our partners are eager to help you with any questions related to the published solutions, as is the GCP sales team. And as always, please don’t hesitate to reach out to us at anthos-onprem-windows@google.com if you have any feedback or need help unblocking your use case.