<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Mathieu Garstecki on Medium]]></title>
        <description><![CDATA[Stories by Mathieu Garstecki on Medium]]></description>
        <link>https://medium.com/@mathieu.garstecki?source=rss-77fc7475af0------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*YsXrd1JL0ONVzcqx6JI45Q.jpeg</url>
            <title>Stories by Mathieu Garstecki on Medium</title>
            <link>https://medium.com/@mathieu.garstecki?source=rss-77fc7475af0------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 12:21:48 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mathieu.garstecki/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[How Back Market SREs prepared for Black Friday]]></title>
            <link>https://engineering.backmarket.com/how-back-market-sres-prepared-for-black-friday-5f017f343408?source=rss-77fc7475af0------2</link>
            <guid isPermaLink="false">https://medium.com/p/5f017f343408</guid>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[gcp]]></category>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[sre]]></category>
            <category><![CDATA[k6]]></category>
            <dc:creator><![CDATA[Mathieu Garstecki]]></dc:creator>
            <pubDate>Thu, 28 Jan 2021 16:57:47 GMT</pubDate>
            <atom:updated>2021-01-28T17:01:01.304Z</atom:updated>
            <content:encoded><![CDATA[<h4>Or the unexpected virtues of load testing</h4><figure><img alt="Mountain-themed rollercoaster carts climbing their initial launch ramp" src="https://cdn-images-1.medium.com/max/1024/0*SZV0DL633KYXAhbT" /><figcaption>Prepare for turbulence. Photo by <a href="https://unsplash.com/photos/E_f6X_kSkp4">Rodolfo Marques</a></figcaption></figure><p><em>Written with </em><a href="https://medium.com/u/9041839b36ec"><em>Theotime Leveque</em></a><em>.</em></p><p>Black Friday (BF in the jargon) is a big event for an e-commerce platform. <a href="https://story.backmarket.com/black-friday-its-dramatic-social-and-environmental-impact/">Whether we like it or not</a>, we’re going to get hammered by customers looking for good deals. At Back Market, we easily get five times the normal traffic during this period, and making sure we’re prepared for the event is always a big part of autumn. With the company’s growth and the constant changes in our architecture, <strong>every Black Friday has been a pain</strong>: new and unexpected issues, plannings blown up by last-minute fixes.</p><p>2019 was a strong vintage in this respect. We had opened several countries since Black Friday 2018, our traffic had steadily grown year-on-year and many components were starting to show signs of struggle. Incidents were increasing in frequency and we were a bit late in bringing those components up to speed to give them room to breathe. With the big day looming and everyone focusing on their own perimeters, an informal poll in the company two weeks before the event made evident that <strong>no one was confident our platform could handle it: big oops</strong>. What ensued were ten days of a frantic war room, with emergency fixes based on everyone’s gut feelings, lots of uncertainty, and a stressful Black Weekend.</p><figure><img alt="Cartoon dog from popular meme extinguishes a fire, shouting “there was no reason to let it last this long and get this bad”." src="https://cdn-images-1.medium.com/max/800/0*ZVwSyIRdsMU16F1p" /><figcaption>Why oh why did we wait for the last minute? <a href="https://www.google.com/url?q=https://twitter.com/pebblepossum/status/1009498357864259584&amp;sa=D&amp;ust=1611776367826000&amp;usg=AOvVaw3QS8INg8vTjcKnhPG-bptQ">Source</a></figcaption></figure><h3>Never again</h3><p>The post-mortem of 2019 left us with a sizable backlog of improvements for 2020. All teams took their share of those improvements in their own plannings and started working on them over the year.</p><p>Fast forward to September 2020. One of the issues identified in 2019 was the <strong>need to have more focus over the year towards resiliency and performance</strong>. While those must remain a concern for every team, some dedicated focus is needed: <strong>enter the SRE team.</strong> Planned early on but created only in September 2020, the goal of our team is to <strong>align Back Market’s resiliency and performance goals with its business stakes,</strong> <strong>ensure every feature team has the right tools to measure and improve its reliability and performance, and to provide expertise on those aspects to everyone.</strong></p><p>With this mission in mind and Black Friday less than three months away, we decided to look for what was needed to reach it smoothly. Reviewing the post-mortem of 2019, a lot of the items have been tackled over the year, our incident frequency have reduced a lot and performance was way better thanks to everyone’s work. Still, informal querying around the company showed worries remained high: was it enough to survive BF? Everyone who experienced last year’s BF remembered the wake-up call right before the event and wanted to avoid that frantic fear-based period, but was <strong>lacking factual data to measure our preparation</strong>. Our short-term objectives quickly became apparent:</p><ul><li>Provide everyone cold, hard data to assess our readiness.</li><li>Structure everyone’s action plan based on that data, and reserve our guts for Aperi-Zooms to celebrate a smooth Black Friday.</li></ul><h3>Load testing: a shortcut of shortcuts</h3><p>Two months, two full-time SREs. That’s all we had to structure the preparation of <em>this buying frenzy event</em>. With little time to act (and react in case of worrying findings), we had to focus on maximum impact with minimum effort¹. Since being overloaded by our traffic was our biggest concern, we decided to <strong>focus primarily on load testing our platform</strong>: checking very practically that we would survive the expected traffic.</p><p>Our objectives were twofold. First, reproduce expected Black Friday traffic (load tests) and verify we will survive. Second, test beyond expected Black Friday traffic (stress tests) to plan for the unexpected, and find the bottlenecks that are likely to fail first if we reach record traffic.</p><p>We had an issue though. As the SRE team was created with two previous members of the infrastructure team, we were aware of the raw limits of the platform and knew we didn’t reach yet a state of the art dev/prod parity… The only place where we would learn enough from load tests was the prod itself… And again, no time to build a viable replica. But what if we break it? Well, the answer was obvious:</p><blockquote><strong>Better break it now so it doesn’t during BF</strong></blockquote><p>Back Market being an impact-driven company, we were confident we could sell this principle to all departments. Still, this being a new and global risk for the platform, we decided to <strong>over-communicate, and present the full detailed plan to every department</strong>, especially business ones. It was a lot of work, but with very useful results: on top of building trust in the plan, it showed that the main worry of our stakeholders was not about breaking the platform, but the impact of the load test traffic on BI data! <strong>Presenting the plan straight to the BI stakeholders gave us this unexpected insight</strong>, and we focused on making the load tests traffic identifiable when designing our scenarios.</p><h3>No free SaaS lunch for you</h3><p>High-value work, little time. When looking at where to cut corners, we saw little value in designing and running our own infrastructure to host our load test clients: we needed to have large capacity immediately, only for a few hours each week, and to have good performance from the start². <strong>Managed hosting was an obvious choice</strong>, and we started benchmarking platforms. We were however quickly disappointed: costs skyrocketed for the load we expected to need or went immediately into private pricing territory which we didn’t have time to negotiate. Developer experience was rarely compelling either: solid API or CLI offerings were not the strong points of most solutions, as was integrating the solutions with our existing DataDog monitoring. So <strong>maybe rolling our own would actually be viable?</strong> A quick POC and back-of-the-envelope cost calculations made us confident we could get it done efficiently on self-hosted GKE clusters.</p><p>We didn’t go as far as creating our own load testing framework. We compared a few tools: <a href="https://gatling.io/">Gatling</a>, <a href="https://locust.io/">Locust</a>, and <a href="https://k6.io/">k6</a>. Gatling is a well-proven, battle-hardened solution, but its <em>indirect</em> way of describing scenarios would have made it painful for our use case. To describe accurately our user journeys, we need to extract content from a page to drive the navigation to the next one. While Gatling provides a way to extract and move data around, it <a href="https://stackoverflow.com/questions/65778356/gatling-correlating-a-json-response-and-modifying-it-in-session">is quite verbose</a> and cumbersome at scale.</p><p>Locust and k6 provide a more direct way to chain calls dynamically in scenarios. For example, for k6:</p><pre>// Load a page<br>let res = http.get(&quot;https://www.backmarket.fr/&quot;);</pre><pre>const doc = parseHTML(res.body);<br>// Get all its &lt;a&gt; tags<br>let links = doc.find(&#39;a&#39;);</pre><pre>links.forEach(function(link) {<br>  // Call the URI for each &lt;a&gt; tag<br>  let subRes = http.get(link.attr(&#39;href&#39;));<br>});</pre><p>In the end, <strong>we went with k6</strong> over Locust: it provides <a href="https://k6.io/docs/results-visualization/datadog">native DataDog integration</a>, with live streaming of metrics and <a href="https://k6.io/docs/using-k6/metrics#custom-metrics">a very easy way to add custom ones</a>, so it was <strong>easier to integrate the tool with our existing monitoring</strong> and have all metrics in a single interface.</p><p>So, GKE and k6. All Google Cloud configuration was done with Terraform, using the very rich <a href="https://github.com/terraform-google-modules">GCP modules for Terraform</a>. To run the tests at scale on the clusters, we went with custom automation. Each user journey is implemented as a k6 scenario (a javascript file), and the load tests are run through Kubernetes Jobs.</p><figure><img alt="How we release our load test scenarios from our GIt repository to our Kubernetes clusters, and how we launch them" src="https://cdn-images-1.medium.com/max/1024/1*P0eCbRsMRgqhYXESowHKbw.png" /></figure><p>All the scenario scripts are uploaded to a GCS bucket on release. For each scenario, the Helm chart deploys as a <a href="https://kubernetes.io/docs/tasks/job/automated-tasks-with-cron-jobs/#suspend">suspended</a> CronJob (so it never runs) in Kubernetes:</p><pre>{{- range $job_name, $job_options := .Values.cronjobs }}<br>apiVersion: batch/v1beta1<br>kind: CronJob<br>metadata:<br> name: loadtest-{{ $job_name }}<br>spec:<br> schedule: &quot;{{ $job_options.schedule }}&quot;<br> suspend: true<br> jobTemplate:<br>   spec:<br>     completions: {{ $job_options.jobsParallelism }}<br>     parallelism: {{ $job_options.jobsParallelism }}<br>     template:<br>       spec:<br>         serviceAccountName: &quot;{{ $.Values.ServiceAccountName }}&quot;<br>         initContainers:<br>         - name: {{ $job_name }}-init<br>           image: &quot;google/cloud-sdk:slim&quot;<br>           command: [&quot;gsutil&quot;, &quot;cp&quot;, &quot;-r&quot;, &quot;{{ $.Values.GCScenariosFullPath }}&quot;, &quot;/&quot;]<br>           resources:<br>             limits:<br>               memory: &quot;2Gi&quot;<br>               cpu: &quot;500m&quot;<br>           volumeMounts:<br>           - mountPath: {{ $.Values.ScenariosVolumeMountPath }}<br>             name: scenarios-cache<br>         containers:<br>         - name: {{ $job_name }}{{ include &quot;test-runner.fullname&quot; $ }}<br>           image: &quot;{{ $.Values.image.repository }}:{{ $.Values.image.tag }}&quot;<br>           command:<br>             - &quot;k6&quot;<br>             - &quot;run&quot;<br>             - &quot;-d&quot;<br>             - &quot;{{ $job_options.maxDuration }}&quot;<br>             - &quot;-i&quot;<br>             - &quot;{{ $job_options.loadIterations }}&quot;<br>             - &quot;-u&quot;<br>             - &quot;{{ $job_options.loadVUs }}&quot;<br>             - &quot;{{ $.Values.ScenariosVolumeMountPath }}/{{ $job_options.scriptPath }}&quot;<br>           env:<br>           - name: K6_DATADOG_HOST<br>             valueFrom:<br>               fieldRef:<br>                 fieldPath: status.hostIP<br>           - name: K6_DATADOG_ADDR<br>             value: &quot;$(K6_DATADOG_HOST):8125&quot;<br>           - name: K6_OUT<br>             value: datadog<br>           volumeMounts:<br>           - mountPath: {{ $.Values.ScenariosVolumeMountPath }}<br>             name: scenarios-cache<br>         volumes:<br>         - name: scenarios-cache<br>           emptyDir: {}<br>---<br>{{- end}}</pre><p>The list of scenarios is described in the Helm chart’s values:</p><pre>cronjobs:<br>  scenario1:<br>    loadIterations: 1600<br>    loadVUs: 20<br>    maxDuration: 45m<br>    jobsParallelism: 10<br>    schedule: 0 8 * * *<br>    scriptPath: scenario1/test.js<br>  scenario2:<br>    loadIterations: 3200<br>    loadVUs: 40<br>    maxDuration: 45m<br>    jobsParallelism: 20<br>    schedule: 0 8 * * *<br>    scriptPath: scenario2/test.js<br>  ...</pre><p>Adding new scenarios or tuning their scale is as easy as changing the values of the Helm chart and redeploying it. The CronJobs never actually run, we just use them as templates. To launch an actual load tests session, we create K8S jobs from all the CronJobs:</p><pre>test_date=$(date +%s)<br>for scenario in $(ls $REPO/scenarios/); do<br>  job_name=loadtest-${scenario}-${test_date}<br>  template_cronjob_name=cronjob/loadtest-${scenario}<br> <br>  kubectl create job ${job_name} --from=${template_cronjob_name}<br>  echo &quot;Launched job ${job_name} based on ${template_cronjob_name}&quot;<br>done</pre><p>While in no way turnkey, <strong>this system was relatively fast to set up</strong> for the first load test sessions: two weeks for two senior engineers, with full Terraform automation, including the setup of dedicated GKE clusters. We’re already a full-Kube company so there was no debate on the platform, but in retrospect, <strong>K8S definitely helped us here</strong>: native Job parallelism and cluster autoscaling allowed us to test early, scale fast when needed, and let the cluster scale back down to limit costs. In the end, it was quite sufficient for our needs and quickly deployed: <strong>SaaS is not always faster!</strong></p><h3>Results</h3><p>Two months of campaign exposed a lot of tech issues of all kinds. It helped to validate and tighten our CDN caching plan, uncovered weaknesses in the scheduling of our Kubernetes clusters when autoscaling, exposed weird CPU throttling behavior in our apps… While most of the work truly <em>had</em> been done over the year, <strong>the load tests still showed a lot of weak points to tighten</strong>, and all our feature teams worked hard on fixing them.</p><p>Beyond the mere tech gains, <strong>the biggest impact was on peoples’ confidence.</strong> The campaign did manage to prove factually that our platform would handle Black Friday smoothly. In 2019, justified fear drove our actions, with a last-minute war room and frantic and risky fixes. <strong>In 2020, it is with justified confidence that we reached BF,</strong> and last-minute fixes were small and targeted. No emergency war room for us!</p><figure><img alt="A calm lake" src="https://cdn-images-1.medium.com/max/1024/0*68fI7BhpqSMKZM3A" /><figcaption>Photo by <a href="https://unsplash.com/photos/R-wqegECgnk">Daniel Bowman</a></figcaption></figure><p>The campaign also helped to put performance and resiliency in everyone’s minds. Two months of dedicated focus, with <strong>performance-reports twice a week</strong> on every team perimeter <strong>helped raise awareness on the complexity of performance issues, and how deeply they touch every aspect of a tech company.</strong> A great prologue to deeper SRE work 😇</p><p>Oh, and how did Black Friday go in the end? Just. F*cking. Perfectly. With all this preparation, it was actually one of the most uneventful periods of the year. <strong>Incidents reported? Zero. Overloaded systems? Zero. </strong>It was fun to be in the war room³ on B-day, stressing out over all the monitoring, and seeing nothing happen… <strong>Which is exactly how it should be!</strong></p><h3>What next?</h3><p>The load tests campaign was the first big action of our fresh SRE team. In two intense months, we managed to <strong>prove that we were doing pretty well in preparation for Black Friday and to largely improve our resiliency in the process to boot.</strong> Still, those two months took their toll and left us quite drained after the event. While we don’t regret doing it, clearly this approach is not sustainable at all: “never again”, again…</p><figure><img alt="An otter sleeping on a rock" src="https://cdn-images-1.medium.com/max/604/0*RpMxmlx3EHEkfZ5c" /><figcaption>Us, two weeks after BF. From <a href="https://imgur.com/gallery/q3a2FtK/comment/35216349?nc=1">Imgur</a></figcaption></figure><p>And so, the real work of the SRE team begins now. Proving in a one-shot campaign that we can survive The Peak is very useful to the company, but <strong>what we truly need is to make sure we offer the best performance and reliability to our users all year long.</strong> We should always be ready for a big peak, we should always know where our weaknesses are, and everyone should always know how to prioritize resiliency/performance versus new features. Our work will be to <strong>empower every team to reach those objectives</strong>: help them refine their SLI/SLO, debug complex transverse issues, attack our platform constantly with chaos engineering, …</p><p>And maybe, in 2021, Black Friday will really become just another Friday?</p><p><strong>If you want to join our Bureau of Technology or any other Back Market department, </strong><a href="https://jobs.lever.co/backmarket"><strong>take a look here, we’re hiring!</strong></a><strong> <em>🦄</em></strong></p><figure><img alt="Back Market logo" src="https://cdn-images-1.medium.com/max/741/0*BahWOzfXrc0OpXwi.gif" /></figure><h3>Footnotes</h3><p>¹ Well, <a href="https://medium.com/back-market-engineering/meeting-philosophy-at-backmarket-c58598343328">even more than usual</a>, in any case.<br>² In our experience, load testing is half about generating load on the target servers, half about fixing bottlenecks in the load test infra itself…<br>³ Yeah, we still had one, because not having one is really taunting Murphy’s Law… But this year it was mostly about watching graphs stay flat and telling anecdotes to pass time.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5f017f343408" width="1" height="1" alt=""><hr><p><a href="https://engineering.backmarket.com/how-back-market-sres-prepared-for-black-friday-5f017f343408">How Back Market SREs prepared for Black Friday</a> was originally published in <a href="https://engineering.backmarket.com/">Back Market Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Paris Container Day 2019]]></title>
            <link>https://engineering.backmarket.com/the-paris-container-day-2019-4db8f7a7ed35?source=rss-77fc7475af0------2</link>
            <guid isPermaLink="false">https://medium.com/p/4db8f7a7ed35</guid>
            <category><![CDATA[container-day]]></category>
            <category><![CDATA[kubernetes]]></category>
            <category><![CDATA[docker]]></category>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[tech]]></category>
            <dc:creator><![CDATA[Mathieu Garstecki]]></dc:creator>
            <pubDate>Tue, 16 Jul 2019 08:16:42 GMT</pubDate>
            <atom:updated>2020-01-07T08:04:43.789Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Written by </em><a href="https://twitter.com/matgarstecki"><em>@matgarstecki</em></a><em>, </em><a href="https://twitter.com/afranche"><em>@afranche</em></a><em> and Aurélien Delobelle.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/533/1*kR5OXU8V4FhtfnG-SXyDRQ.jpeg" /></figure><p>In the Spring, container conferences sprout everywhere! One was the <a href="https://paris-container-day.fr/">Paris Container Day</a> in June, which we had the chance to attend. With one day and around 20 talks, it was a more modest event than behemoths like KubeCon (which we’ll talk about soon), and focused mostly on container platforms rather than on developing containers.</p><p>The publicized theme of the conference was “Standards &amp; Craftsmanship”. Rehashing the parallel with shipping containers as “the standard that led to globalized trade,” standardization efforts and their effects were indeed well covered.</p><h3>Security in containers</h3><p>With Aqua and CyberArk as sponsors, it was clear that there would be an emphasis on the topic of security in the “containerization” process and orchestration at the event. Sure enough, <a href="https://www.youtube.com/watch?v=_l56oUxHSio"><em>The State of Kubernetes Security</em></a> was one of the morning’s keynotes, serving to introduce the rest of the conference.</p><p>Under the headline of Standardization, we were reminded of standard bodies like the <a href="https://learn.cisecurity.org/benchmarks">CIS</a> that publish security benchmarks and establish a number of safety rules, including those for Kubernetes. More practically, they introduced us to <a href="https://github.com/aquasecurity/kube-bench">kube-bench</a>, a tool offered by Aqua Security to help ensure that a Kubernetes setup meets these security benchmarks. They also covered breach examples (i.e., <a href="https://www.wired.com/story/cryptojacking-tesla-amazon-cloud/">the known breach that happened at Tesla</a>). The takeaway was that it was important not to forget your basics: limiting permissions using RBAC, updating your clusters, etc.</p><p>Down below, in the kernel itself, eBPF is rapidly emerging as the new network and observability standard of choice. It provides a <a href="https://www.youtube.com/watch?v=S0zZ8nIlxkM">generic language to hook and analyze any event in the kernel</a>, and after five years of development, it seems mature enough to underpin the next generation of networking and security frameworks.</p><p>Falco uses it to <a href="https://youtu.be/XIIBwz-wtIs?t=1442">detect misbehaving containers</a> easily by hooking onto every kernel event of the container, and would probably be prohibitively slow without the in-kernel decisions that eBPF provides. On the network side, Cilium aims to <a href="https://www.youtube.com/watch?v=FsrZZ0nVrWU">implement a very efficient routing and firewalling layer</a> for container platforms, again leveraging eBPF to tag, track, and filter packets in a very detailed way. Cilium’s capacities reach almost into the domain of service meshes, and it looks very promising to boost network performance and security in Kubernetes.</p><h3>Standards make room for new standards</h3><p>These past few years, Kubernetes has become the reference for container hosting, and several speakers at the conference noted how the <a href="https://github.com/ramitsurana/awesome-kubernetes#managed-kubernetes">offers for managed Kubernetes</a> have multiplied, with most infrastructure providers having some kind of Kubernetes hosting. This de-facto standard allows richer components to abstract themselves from the underlying infrastructure (cloud vs. on-premises), and evolve faster by targeting a single API. Several talks looked at its extensibility.</p><p>For OPS people, the REST API standardizes the description of infrastructure resources (workloads, config, …) in a common format, <a href="https://youtu.be/oYcGzEo15n4?t=724">making many platform-management tasks easier</a>. <a href="https://www.youtube.com/watch?time_continue=1745&amp;v=oYcGzEo15n4">Helm</a> is emerging as the almost-standard package manager for Kubernetes and makes installing complex Kubernetes deployments easy and uniform. The <a href="https://github.com/helm/charts/tree/master/stable">package list</a> covers many applications, like traditional monitoring tools, storage solutions, and even games! At a higher level, Spinnaker is a platform dedicated to executing Continuous Deployment pipelines that integrate Red/Black and Canary Deployments and directly target high-level cloud management APIs. For example, Pivotal showed the integration of Spinnaker to their managed Kubernetes offering, PKS. One can also list Jenkins X, a new variant of Jenkins fully dedicated to Kubernetes CI/CD.</p><p>Beyond the building blocks provided by Kube (pod, deployment, etc.) it is possible to add new primitives to describe your own abstractions. Some of them can also be managed by the Kubernetes API. This was what Bastien Cadiot demonstrated with <a href="https://youtu.be/J0GtDbYCGEo">Knative</a>: this project used the extensibility of Kubernetes to provide three rich new abstractions. Knative Serving allows serverless hosting, similar to AWS Lambda. Knative Build can be used to launch event-triggered pipelines. The last part is Knative Eventing, that allows reacting to events. The joint use of these three elements allow you to react to a notification from your Git hosting (Eventing) and launch a pipeline (Build) whose steps will be executed by a function (Serving).</p><p>Many of the above are implemented using Kubernetes custom resources and the operator pattern, and many more exist. You can find a partial view of these reusable operators <a href="https://github.com/operator-framework/awesome-operators">here</a>.</p><h3>One API to unite them all</h3><p>Kubernetes, Kubernetes, Kubernetes… If everything in this article revolves around it, it’s because nearly every presentation at the Paris Container Day talked about it! To summarize, here’s what we got out of this day.</p><p>Having a common abstraction like Kubernetes frees editors and developers to focus on high-level features, like CI/CD or serverless offerings with a lot of the low-level work (high-availability, networking, scheduling, …) standardized.</p><p>Standardization doesn’t mean the platform has stopped evolving though. Security standards are still being refined, and eBPF promises a shift in the networking and security approaches that should push Kubernetes even further.</p><p>Cilium looks particularly promising on the network side: its promise of a blend between network and service mesh along with great performance are very exciting and we will keep a close watch on its evolution.</p><p>It was a great day overall, and we’re excited to see what the future of cloud-native orchestration will bring with the fast-paced evolution we’re seeing.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4db8f7a7ed35" width="1" height="1" alt=""><hr><p><a href="https://engineering.backmarket.com/the-paris-container-day-2019-4db8f7a7ed35">The Paris Container Day 2019</a> was originally published in <a href="https://engineering.backmarket.com/">Back Market Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>