<?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 Marco Palmisano on Medium]]></title>
        <description><![CDATA[Stories by Marco Palmisano on Medium]]></description>
        <link>https://medium.com/@palmima8?source=rss-8a5eafb53cbb------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*OOU8yxwp2U0GuSDK</url>
            <title>Stories by Marco Palmisano on Medium</title>
            <link>https://medium.com/@palmima8?source=rss-8a5eafb53cbb------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 20:41:16 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@palmima8/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[Story from Modern Sparta: Are Google DataStream and Oracle RAC SCAN friends?]]></title>
            <link>https://medium.com/edreams-odigeo-tech/story-from-modern-sparta-are-google-datastream-and-oracle-rac-scan-friends-d786198c7bcb?source=rss-8a5eafb53cbb------2</link>
            <guid isPermaLink="false">https://medium.com/p/d786198c7bcb</guid>
            <category><![CDATA[datastream]]></category>
            <category><![CDATA[oracle]]></category>
            <category><![CDATA[rac]]></category>
            <category><![CDATA[scan]]></category>
            <category><![CDATA[google]]></category>
            <dc:creator><![CDATA[Marco Palmisano]]></dc:creator>
            <pubDate>Thu, 20 Mar 2025 12:42:12 GMT</pubDate>
            <atom:updated>2025-03-20T12:42:12.497Z</atom:updated>
            <content:encoded><![CDATA[<h3>The magic world of incompatibilities that become opportunities</h3><p>Welcome to this story from the Modern Sparta!</p><p>I’m Marco, your guide in this fairy world of incompatibilities that magically become opportunities…</p><p>Once upon a time, there was a young Platform Engineer who had been assigned the unfortunate task of creating a Google Datastream stream that could read the data from his reign databases: two majestic Oracle Cloud Exadata, source of the wisdom of all humanity.</p><p>He was starting the day with an excessive optimism that disappeared the instant he read the now well-known oracle (pun intended):</p><pre>Datastream doesn&#39;t support direct connectivity to databases using<br>the Single Client Access Name (SCAN) feature in Oracle Real Application<br>Clusters (RAC) environments. To overcome this limitation, connect directly<br>to one of the nodes, or use the Oracle Connection Manager.<br><br>Source: https://cloud.google.com/datastream/docs/faq#oracle-source</pre><p>At this moment, all the good intentions the poor guy had shattered in his eyes. The reign required a new virtual machine since the CMAN (the aforementioned Oracle Connection Manager) isn’t available as a managed service in OCI.</p><p>A new virtual machine meant extra monitoring, scaling, HA, design to do, approvals to get… and the poor engineer had already lost his optimism: how could he do?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*ZYI1ItMG76b3WeQNPq-TEg.png" /></figure><h3>How many of you have breakfast with napalm?</h3><p>It was a cold and rainy night… ok, let’s stop here and get back to reality.</p><p>We already had Datastream in place at the time, with one connection profile for the DB node.</p><p>Considering that we have two main Exadata Clusters with 2 different databases running inside of them, it meant that we had <strong>eight</strong> connection profiles available to choose:</p><ul><li>ENV01 NODE A DB1</li><li>ENV01 NODE B DB1</li><li>ENV02 NODE A DB1</li><li>ENV02 NODE B DB1</li><li>ENV01 NODE A DB2</li><li>ENV01 NODE B DB2</li><li>ENV02 NODE A DB2</li><li>ENV02 NODE B DB2</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yWm-6jOsipPIgFtg1rYTeQ.png" /></figure><p>Each time we did a switchover (i.e. for maintenance purposes), we needed to <strong>stop, delete and recreate each</strong> <strong>stream</strong>, pointing to the correct cluster, and <em>hopefully</em> the correct node.</p><p>While the number of streams was rising, the problems started peeking: excessive node unbalance, small data loss in the stream because of temporary failures, and manual work in recreating/balancing streams… we needed to find a solution.</p><p>Reading the documentation again led to an interesting idea, let’s go through it together:</p><pre>To overcome this limitation, [...TLDR: CMAN...] <strong>You can also create a private <br>connectivity configuration using reverse proxy solutions</strong> like HAProxy.</pre><p>Hey, stop for a minute: we <strong>do have</strong> managed reverse proxies in OCI! 🤩</p><h3>Failing fast: sleeves up and let’s do a POC</h3><p>I was excited with this new discovery and I started instantly working on that. A little bit of terraform and the work was done: we had a new shiny network load balancer provisioned with all the nodes as targets and TCP health checks on the DB port.</p><p>We’re ready to go!</p><p>At the very beginning (and to our big surprise) it worked fine, but after a couple of minutes the streams started failing randomly with unexpected errors that sounded like <strong>ORA-12154</strong>: TNS:could not resolve the connect identifier specified</p><p>In the meanwhile I was reflecting on my life choices, I started digging a little bit more into the Oracle documentation of SCAN.</p><h3>RTFM: <strong><em>aka</em></strong> “Can you please properly read the documentation?”</h3><blockquote><em>Oracle Database clients connect to the database using a Single Client Access Name (SCAN). […] The SCAN resolves to multiple IP addresses reflecting multiple listeners in the cluster that handle public client connections. When a client submits a request, the SCAN listener listening on a SCAN IP address and the SCAN port is made available to a client. Because all services on the cluster are registered with the SCAN listener, the SCAN listener replies with the address of the local listener on the least-loaded node where the service is currently being offered. Finally, the client establishes connection to the service through the listener on the node where service is offered. All of these actions take place transparently to the client without any explicit configuration required in the client.</em></blockquote><blockquote>Source: <a href="https://docs.oracle.com/en/database/oracle/oracle-database/19/cwlin/about-the-scan.html">https://docs.oracle.com/en/database/oracle/oracle-database/19/cwlin/about-the-scan.html</a></blockquote><p>So, this means that the SCAN protocol involves some kind of intrinsic cleverness: it returns the VIPs of the nodes that actually have the specific service started and running! But how can we reproduce that by using only HTTP or TCP simple health checks?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vKISLPJhxaqFW1_GQfl_KQ.png" /><figcaption>Actual representation of the situation (Credits: Imagen 3 — Gemini Advanced 2.0 Flash)</figcaption></figure><h3>“I am about to do something very bold in this job that I’ve never done before: try.” (Jim Halpert)</h3><p>Considering the limited possibilities that we had in the NLB health check, we started drawing down some ideas: we needed a clearer path to follow.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2eQRFVTARjULzYhyOPNJSA.png" /></figure><p>This seemed our ideal target situation: a lot easier to manage, and we like it! But how can we implement a similar protocol without touching the inner of the database engine?</p><p>In the meantime, I was doing a lot of different tests to try to exploit the protocol in some way, and I noticed that Oracle returned two different errors in this case:</p><ul><li>the service is not started</li><li>the service is started but you’re using the wrong username/password combination</li></ul><p>Bingo! it means that we can spot where the service is running simply by comparing the error code 🥳</p><p>But, wait… how could we do it with a NLB Health Check? ☹️</p><h3>A Gopher for Friend</h3><p>We needed to have <em>something</em> in order for the load balancer to check if the specific node had the DataStream service started, easy peasy.</p><p>Let’s start with the problems:</p><ul><li>We had to have this system in all the different databases and clusters, so we couldn’t hardcode anything in the code for querying the database</li><li>We couldn’t use the SCAN addresses because we needed to know if the service was running in the requested node</li><li>We couldn’t deploy the health check outside of the servers because the Load Balancer can execute the health check only on the registered targets IPs</li><li>We didn’t have advanced options for tuning the health check request</li></ul><p>So, how did we accomplish that?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*009Dg2KayO62QYYxkwgb5w.png" /></figure><p>We decided to create a <a href="https://stackoverflow.com/questions/41024771/micro-service-vs-nano-service">nanoservice</a> that could help us have a smart health check, and after abstracting a little bit the possible configurations and playing with the health check parameters, we reached a promising conclusion: since we had one NLB per database, we could include the check settings in the API call parameter, making the service super easy to deploy without any additional configuration required!</p><p>In particular, we could easily extract:</p><ul><li>Destination server IP from the Host header (Oracle database services listens for connections on VIPs)</li><li>Database name to check (as a path parameter)</li><li>Service name to check (as a path parameter)</li></ul><p>This, combined with the error codes, actually made it possible for us to solve the problem!</p><pre>// main.go<br>import (<br>// ...<br>	go_ora &quot;github.com/sijms/go-ora/v2&quot;<br>)<br><br>// ...<br>func main() {<br> r := mux.NewRouter()<br> r.HandleFunc(**&quot;/checkservicestatus/{db_name}/{svc_name}&quot;**, checks.CheckOracleDatabase)<br> http.Handle(&quot;/&quot;, r)<br>  // ...<br> err := http.ListenAndServe(fmt.Sprintf(&quot;:%s&quot;, environment.GetEnvVar(&quot;ORU_PORT&quot;)), r)<br> if err != nil {<br>  log.Fatal(&quot;ListenAndServe: &quot;, err)<br> }<br> log.Println(&quot;Server started on port:&quot;, environment.GetEnvVar(&quot;ORU_PORT&quot;))<br>}<br>// ...</pre><pre>// checks/oracle.go<br>package checks<br>func CheckOracleDatabase(w http.ResponseWriter, r *http.Request) {<br> vars := mux.Vars(r)<br> hostname := getHostIp(r.Host)<br> if hostname == &quot;&quot; {<br>  w.WriteHeader(http.StatusBadRequest)<br>  return<br> }<br> log.Printf(&quot;hostname: %v\\n&quot;, hostname)<br> log.Printf(&quot;db_name: %v\\n&quot;, vars[&quot;db_name&quot;])<br> log.Printf(&quot;svc_name: %v\\n&quot;, vars[&quot;svc_name&quot;])<br> db_port, _ := strconv.Atoi(environment.GetEnvVar(&quot;ORU_DB_PORT&quot;))<br> // Yes, we&#39;re using &quot;username&quot; and &quot;password&quot; as respectively username and password :)<br> connStr := go_ora.BuildUrl(hostname, db_port, vars[&quot;svc_name&quot;], &quot;username&quot;, &quot;password&quot;, nil)<br> log.Printf(&quot;Generated connection string [%s]&quot;, connStr)<br> conn, err := go_ora.NewConnection(connStr, &amp;configurations.ConnectionConfig{<br>  DatabaseInfo: configurations.DatabaseInfo{<br>   DBName: vars[&quot;db_name&quot;],<br>   SID:    vars[&quot;svc_name&quot;],<br>  },<br> })<br> if err != nil {<br>  log.Println(err)<br>  w.WriteHeader(http.StatusInternalServerError)<br>  return<br> }<br> // More than 5 seconds means something&#39;s wrong, timing out...<br> ctx, cancel := context.WithTimeout(context.TODO(), 5*time.Second)<br> defer cancel()<br> err = conn.OpenWithContext(ctx)<br> if err != nil {<br>  log.Println(err)<br>  _, isCastOk := err.(*network.OracleError)<br>  if !isCastOk {<br>   w.WriteHeader(http.StatusInternalServerError)<br>   return<br>  }<br>  // ORA-1017: Service started but wrong user/password combination<br>  if err.(*network.OracleError).ErrCode == 1017 {<br>   w.WriteHeader(http.StatusOK)<br>   return<br>  }<br>  // ORA-12514: Service not started<br>  if err.(*network.OracleError).ErrCode == 12514 {<br>   w.WriteHeader(http.StatusNotFound)<br>   return<br>  }<br>  w.WriteHeader(http.StatusInternalServerError)<br>  return<br> }<br> w.WriteHeader(http.StatusOK)<br>}</pre><p>This allows us to configure endpoints on the different NLBs easily and using similar formats, using the same code for every deployment we need to have</p><pre>// NLB DB1<br>health_check_protocol         = &quot;HTTP&quot;<br>health_check_port             = &quot;8080&quot;<br>health_check_timeout_in_ms    = 6000<br>health_check_interval_in_ms   = 150000<br>health_check_retries          = 1<br>health_check_http_url_path    = &quot;/checkservicestatus/mysuperdatabase/svc_datastream_flights_db&quot;<br>health_check_http_return_code = 200<br><br>// NLB DB2<br>health_check_protocol         = &quot;HTTP&quot;<br>health_check_port             = &quot;8080&quot;<br>health_check_timeout_in_ms    = 6000<br>health_check_interval_in_ms   = 150000<br>health_check_retries          = 1<br>health_check_http_url_path    = &quot;/checkservicestatus/myamazingseconddatabase/svc_datastream_hotels_db&quot;<br>health_check_http_return_code = 200</pre><h3>Okay, sure, but is this really working?</h3><pre>$ systemctl status ds-oru<br><br>ds-oru.service - Data Spartans Oracle REST Utilities<br>   Loaded: loaded (/etc/systemd/system/ds-oru.service; enabled; vendor preset: disabled)<br>   Active: active (running) since Wed 2024-11-13 09:17:44 UTC; 3 months 20 days ago<br> Main PID: 16405 (ds-oru)<br>    Tasks: 23 (limit: 319999)<br>   Memory: 15.9M<br>   CGroup: /system.slice/ds-oru.service<br>           └─16405 /usr/local/bin/ds-oru</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EX0-S-nJlnB1yDBcwducwg.png" /><figcaption>Datadog monitors for NLBs</figcaption></figure><p>At the time of writing this system has been running for <strong>more than</strong> <strong>8 months</strong> and it helped us survive without incidents of more than 2 switchovers, transient failures and maintenance consuming only a few MB of RAM!</p><p>The yellow lines in the graphs are maintenance events that in the past matched incidents and manual stream recreation 🫣</p><p>As Bill Gates stated:</p><blockquote><strong>“I choose a lazy person to do a hard job</strong><em>. Because a lazy person will find an easy way to do it.”</em></blockquote><p>This is the story of how we avoided creating and managing new VMs to scale, audit, monitor and comfort when feeling sad.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/404/0*uZ7nZXnMXb9bCCOC.png" /><figcaption><a href="https://xkcd.com/1319/"><em>https://xkcd.com/1319/</em></a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d786198c7bcb" width="1" height="1" alt=""><hr><p><a href="https://medium.com/edreams-odigeo-tech/story-from-modern-sparta-are-google-datastream-and-oracle-rac-scan-friends-d786198c7bcb">Story from Modern Sparta: Are Google DataStream and Oracle RAC SCAN friends?</a> was originally published in <a href="https://medium.com/edreams-odigeo-tech">eDreams ODIGEO</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Story from Modern Sparta: How can database user management become easy?]]></title>
            <link>https://medium.com/edreams-odigeo-tech/story-from-modern-sparta-how-can-database-user-management-become-easy-12fc55a2ca98?source=rss-8a5eafb53cbb------2</link>
            <guid isPermaLink="false">https://medium.com/p/12fc55a2ca98</guid>
            <category><![CDATA[database]]></category>
            <dc:creator><![CDATA[Marco Palmisano]]></dc:creator>
            <pubDate>Mon, 24 Feb 2025 14:13:50 GMT</pubDate>
            <atom:updated>2025-02-24T14:13:50.162Z</atom:updated>
            <content:encoded><![CDATA[<h3>The importance of being brave</h3><p>As the ancient Spartans taught us, we never should feel really comfortable: comfort is the main enemy of evolution.</p><p>That means we should always thrive on efficiency, rely on more reliable automation and concentrate our efforts in important tasks.</p><p>So, that’s the story:</p><p>Every one of you who had to deal with IAM knows how painful it is: you need updated information, security agreements, ensure a good developer experience (raise your hand if you were logged out during a night intervention) and a lot of other hidden tasks and meetings that come up regularly.</p><p>Time goes by, and you start to be used to waiting for days, doing repetitive tasks while saying in your mind “Yeah it happens only once a month” and in the end spend a lot of “5 minutes” during your day.</p><p>As said before, we need to avoid comfort, so we decided to eradicate the problem: <strong>once and for all</strong></p><p><a href="https://media4.giphy.com/media/v1.Y2lkPTc5NDFmZGM2ZGp6a3Bwdm44MWI4ZjJkdDVjdGVicnhneXVvOWg5bWRzMGJhcHlmMSZlcD12MV9naWZzX3NlYXJjaCZjdD1n/bh5UZVQjB9ouQ/giphy.gif">satan GIF - Find &amp; Share on GIPHY</a></p><p>OK, let’s start with the usual trip: we collected feedback on the current process, discussed some ideas, and thought about a small solution that could help us be <em>only</em> faster.</p><h3>Need for Speed</h3><p>The first time I heard the phrase “9 women can’t give birth in 1 month” I felt confused, but if you let it settle inside of you it becomes like a little funny mantra.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*YLGK2nsWEn1qILXD" /><figcaption>Photo by <a href="https://unsplash.com/@leo_visions_?utm_source=medium&amp;utm_medium=referral">Leo_Visions</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>In an ideal world (figure above), we would have all the needed time to study the perfect solution, but like a friend of mine likes to say “Software shouldn’t be built following the rules of an ideal world”.</p><p>CloudSQL Postgres instances in the company were starting to pop up frequently, and the amount of IAM tickets was growing at a constant rate:</p><ul><li>we needed to be fast and effective</li><li>the product needed to be fast and easy to use</li></ul><h3>Fail fast</h3><p>We started from the very basics: agreements.</p><p>We took care of having meetings and agreements with all the stakeholders:</p><ul><li>Security Office, which ended up happy with their new shiny expiration policies implemented</li><li>Data Experience Team, that wouldn’t need to manage IAM tickets anymore</li><li>Developers started to have access in hours instead of days</li></ul><p>Although I also spent a too much time discussing with Copilot who of us writes better code, we wrote down a roadmap for the tool and started to iterate</p><h3>SDUM!</h3><p>SDUM is the acronym for Smart Database User Management, and this is the story of how we created the tool that saved us hours of manual work.</p><h3>Let’s start from the beginning: what is what</h3><p>That’s what we needed: a Python CLI for managing permissions in a GitOps-compliant way!</p><p>SDUM persists in its state in Git repositories, and it’s the owner of the reconciling cycle that ensures only the defined permissions in its state are effectively present in the databases.</p><p><em>If you’re thinking in Terraform, yes: you’re thinking well.</em></p><h3>Ok, you got me: how it works?</h3><p>SDUM runs on simple CI pipelines, triggered by three possible events:</p><ul><li>Cron schedule: it runs to ensure expired users are removed from the database and are not re-added manually</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/498/0*_YPAUup6UWsMVMi7.png" /></figure><ul><li>Git commit: data operators can manually commit to the repository instead of manually logging in to database servers and executing the CREATE USER, ALTER USER commands</li><li><strong>Backstage interaction:</strong> we decided to eliminate completely the manual process! Now the Lead Engineer of each team can manage permissions on its own, without the need to open a ticket and wait days <br>(The backstage plugin calls GitLab APIs under the hood for updating the state repository.)</li></ul><h3>That’s fun: but, what kind of permissions do you assign?</h3><p>We worked hard for months before letting development teams be independent in their database creation, the majority of the time was spent in defining conventions and standards: this definitely paid off in the long term.</p><p>SDUM was built around this company conventions during its first iteration (remember, fail fast!) and to be honest, it’s working better than expected!</p><p>It persists its state in simple YAML files, easy to modify programmatically and manually:</p><pre>instance: the.instance.secret.dns.recipe</pre><pre>permissions:<br>- user: superman@edreamsodigeo.com<br>	role: readonly_database_role<br>	expiration_date: 2025-01-20<br>...</pre><p>Yes, it’s <strong>this</strong> easy.</p><p>Keeping the file format both human and machine-readable sped up a lot the following steps of the project, so please remember to think twice about your data structures 🫰</p><p>We considered also hidden fields that can be used to override some tool behaviours avoiding then to have hardcoded strategies, and surprisingly, we discovered the magic of defaults and overrides. /s</p><h3>CloudSQL IAM</h3><p>Well, Kudos to Google for having implemented this magnificent piece of code that allows us to be compliant and safe without giving up on daily experience!</p><p>As you may have spotted in the .sdum file example, we leverage existing IAM identities for logging in to databases, making user management easier on every level 🚀</p><h3>The time management</h3><p>Thanks to my coding mentors, my OCDs now feature among them the idea of being fast and precise, ninja style 🥷</p><p>In order to be fast we decided to split the project into 3 milestones that we had to deliver in 3 months:</p><ul><li>Documentation</li><li>Internal Use (Git)</li><li>Good-looking UI for everybody</li></ul><p>We focused from the very first moment (sometimes forcing ourselves) on the project documentation: we wouldn’t have time to document later so we needed to do both at the same time.</p><p>Among the boring documentation came the PoC: I had never written a complete project in Python starting from scratch, so I wanted to test out if our beautiful paper diagrams could make sense in real life (spoiler: they didn’t)</p><p>We decided to implement a simple and clear folder structure that could allow us to be fast and expand the tool in the future.</p><pre>src/<br>|______ core/<br>|______ services/<br>|______ base_database_service.py<br>    |______ state/<br>            |______ base_state_class.py<br>            |______ base_state_handler.py<br>    |______ auditing/<br>            |______ base_logger.py<br>|____  commands/<br>    |______ validate/<br>    |______ plan/<br>    |______ apply/<br>    |______ base_command.py<br>|____  utils/<br>...</pre><p>After validating the base cases and some corner implementation, we continued evolving the product based on the feedback we received from our first stakeholders, our data teams!</p><p>This allowed us to create a robust foundation for our core that ensured coverage for the corner cases we found while testing in production.</p><p>The moral here is simple: don’t focus too much on abstraction and don’t do chatgpt_gave_me_this_semms_to_work_v1.py, “perfection” is always a question of balance.</p><h3>Backstage integration</h3><p>As told at the beginning, we needed to offload this work to someone.</p><p>Who better of the requesters themself? :D</p><p>Since different years have passed using Backstage in our company, it holds all the information about guess what: teams, people, and applications. It was only a matter of integrating our work, a couple of days and it’s done! (never did a worst estimation in my life)</p><p>The final integration with Backstage consisted of 3 different plugins:</p><ul><li>A front-end plugin, that holds the UI code used by Data operators and Lead Engineers to commit safely in the state repositories respecting the standards</li><li>A back-end plugin, that is capable of rendering valid .sdum files and sending them to our VCS</li><li>A catalogue plugin, in order to let Backstage ingest our custom resources</li></ul><p>After hours of debugging and testing (thanks Typescript, I remembered why I used to love you), we asked our first “customers” to test out the new shiny product, and I can’t tell you how much we felt relieved, knowing that we wouldn’t have received Database Access Control JIRA tickets anymore.</p><h3>What have we learned from this?</h3><p>After several months spent in this personal initiative, I can share a secret with you: while I thought the satisfaction would have come from the productivity increase, the real benefit I felt was seeing other teams influenced by this.</p><p>Don’t get me wrong, we already have a lot of engineering and automation in place, but remember that the majority of the tools we all use daily (K8s, ArgoCD, Helm, Prometheus) were born as internal projects, and if I get even one of you readers developing a new groundbreaking tool it’s already enough for me :)</p><h3>Can you please finally explain to me what is SDUM?</h3><p>Yes, sure.</p><p>SDUM is the story that tells how we saved more than <strong>5 months</strong> of manual work <strong>every year</strong> investing only 3 months of passionate work (and this being independent from the number of instances that will be created).</p><p>Someone once said that if a task needs to be run more than 3 times, it’s worth automating it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/550/0*wCkhh152B69kwnf4.png" /><figcaption><a href="https://xkcd.com/974/">https://xkcd.com/974/</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=12fc55a2ca98" width="1" height="1" alt=""><hr><p><a href="https://medium.com/edreams-odigeo-tech/story-from-modern-sparta-how-can-database-user-management-become-easy-12fc55a2ca98">Story from Modern Sparta: How can database user management become easy?</a> was originally published in <a href="https://medium.com/edreams-odigeo-tech">eDreams ODIGEO</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>