BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//saas-killa//agenda//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:AI Engineer
BEGIN:VEVENT
UID:c919f12e-7657-4cdf-83b4-c7d8cf69bf84@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T090000Z
DTEND:20261110T094500Z
SUMMARY:Gray Failure: The Achilles’ Heel of Cloud-Scale Systems
SEQUENCE:1
LOCATION:Main hall
DESCRIPTION:Tvorimir Luchakivskiy\nOps · Talk (45 min)\n\nCloud scale prov
 ides the vast resources necessary to repair and fix failed components\, bu
 t this is useful only if those failures can be detected. For this reason\,
  the major availability breakdowns and performance anomalies we see in clo
 ud environments tend to be caused by subtle underlying faults\, i.e.\, gra
 y failure rather than fail-stop failure. In this talk\, we discuss our exp
 eriences with gray failure in Microsoft Azure to show its broad scope and 
 consequences with several case studies. We also argue that a key feature o
 f gray failure is differential observability: that the system’s failure 
 detectors may not notice problems even when applications are afflicted by 
 them. We will show how Microsoft Azure applied the differential observabil
 ity in practice and bridged the gap between differe
END:VEVENT
BEGIN:VEVENT
UID:f372e033-425b-46dc-b666-833deab908fe@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T090000Z
DTEND:20261110T094500Z
SUMMARY:Real Talk: What We Think We Know — That Just Ain’t So
SEQUENCE:1
LOCATION:Studio
DESCRIPTION:Marcus Mccoy\nPractice · Talk (25 min)\n\nDr. David Woods once
  said to me “We cannot call it a scientific field unless we can admit we
 ’ve gotten things wrong in the past.” Do we\, in this community\, do t
 hat? Well-formed critique is critical for any field — including SRE — 
 to progress. I’d like to talk about a few ideas\, assumptions\, and conc
 epts often talked about in this community but whose validity is rarely que
 stioned or explored. John Allspaw has worked in software systems engineeri
 ng and operations for over twenty years in many different environments. Jo
 hn’s publications include the books *The Art of Capacity Planning* (2009
 ) and *Web Operations* (2010) as well as the forward to *The DevOps Handbo
 ok.* His 2009 Velocity talk with Paul Hammond\, *10+ Deploys Per Day: Dev 
 and Ops Cooperation* helped start the DevOps movement. John
END:VEVENT
BEGIN:VEVENT
UID:96039f64-2d32-44a5-a0b8-1b39fb393ae3@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T090000Z
DTEND:20261110T094500Z
SUMMARY:Product Reliability for Google Maps
SEQUENCE:1
LOCATION:Workshop room
DESCRIPTION:Nora Leroy\nResearch · Talk (25 min)\n\nAs our organization ha
 s gotten very good at protecting server SLOs with reliability best practic
 es like scaling globally distributed at-scale architectures\, toil mitigat
 ion\, and continuous reliability improvements we noticed that a majority o
 f incidents impacting our end-users were not showing up as an SLO miss. In
  many cases these outages were not even observable from the server side - 
 for example\, the rollout of a new version of the consumer mobile applicat
 ion (that our services powers) to an app store could break one or more cri
 tical feature(s) due to bugs in client code. This reality has led to a cha
 nge in the way we approach reliability - we’re shifting our focus from s
 erver reliability to product reliability. We’re not yet finished with th
 e transition\, but we’re starting to see very 
END:VEVENT
BEGIN:VEVENT
UID:c0e25874-8dd7-4fa6-be8e-128f4e117872@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T100000Z
DTEND:20261110T104500Z
SUMMARY:Using Generative AI Patterns for Better Observability
SEQUENCE:1
LOCATION:Main hall
DESCRIPTION:Sidónio Moreira\nSystems · Talk (25 min)\n\nWhat happens when
  you've got a team of SREs who get downsized and smacked with an increasin
 g workload? This is the story of my small team's wild ride of revamping ou
 r in-house incident tool. Spoiler: It turned into a "make or buy" debate r
 eal quick! Picture this: a team of 10 becomes a merry band of 3\, drowning
  in a sea of issues\, requests\, and feedback. Unexpectedly\, our team cha
 nged like a game of musical chairs\, skills shuffling all over. In our pur
 suit to be the heroes\, we hit the brakes and pitched buying a ready-made 
 tool for a change. It wasn't just a business decision\; it was a cultural 
 rollercoaster for everyone involved! So\, let's take a stroll down memory 
 lane\, exploring how we switched gears to evaluate third-party tools to pa
 ve the ground for improving the reliability of our p
END:VEVENT
BEGIN:VEVENT
UID:b1147d49-190c-4d37-98e9-294b0f200fc0@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T100000Z
DTEND:20261110T104500Z
SUMMARY:Strengthening Apache Pinot's Query Processing Engine with Adaptive 
 Server Selection and Runtime\n## Query Killing
SEQUENCE:1
LOCATION:Studio
DESCRIPTION:Irmingard Frese\nPractice · Talk (45 min)\n\nApache Pinot\, a 
 real-time\, distributed OLAP database\, encountered resiliency challenges 
 in its large production deployment due to server failures\, slowness\, and
  unpredictable query patterns. To ensure adherence to strict SLAs\, enhanc
 ements were made to its Query Processing framework through two key feature
 s: Intelligent Query Routing and Runtime Query Killing. Intelligent Query 
 Routing replaces the traditional round-robin server query distribution wit
 h a dynamic approach based on continuous server performance metrics. This 
 adaptive method proactively routes queries to more efficient servers\, red
 ucing the need for manual intervention and preserving availability during 
 transient server issues. Runtime Query Killing addresses the issue of comp
 lex queries causing Out of Memory errors and avail
END:VEVENT
BEGIN:VEVENT
UID:8017c278-73a1-4f86-b923-c0f8e45fcdc9@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T100000Z
DTEND:20261110T104500Z
SUMMARY:Optimizing Resilience and Availability by Migrating from JupyterHub
  to the Kubeflow Notebook\n## Controller
SEQUENCE:1
LOCATION:Workshop room
DESCRIPTION:Andrea Johnson\nResearch · Talk (45 min)\n\nThis presentation 
 details our transition from JupyterHub to the Kubeflow Notebook Controller
 . JupyterHub was architected in a backend agnostic way that "supports" Kub
 ernetes but isn't truly Kubernetes-native. As a result\, it has significan
 t shortcomings with respect to resilience and high availability. In partic
 ular\, the core component\, the hub API\, can only have one replica at any
  given time. In contrast\, The Kubeflow Notebook controller is built from 
 the ground up to be Kubernetes native using the operator pattern. There's 
 far less complexity\, fewer components\, less brittleness\, and improved r
 esilience and high availability. As a result\, our platform has been able 
 to scale to four times as many users\, including ten times as many concurr
 ent executions. Our users are happier and there's le
END:VEVENT
BEGIN:VEVENT
UID:079ab51f-d8d9-401e-b30d-39b20f5bd54f@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T110000Z
DTEND:20261110T114500Z
SUMMARY:Cross-System Interaction Failures: Don't Fail through the Cracks
SEQUENCE:1
LOCATION:Main hall
DESCRIPTION:Larisa Kladochniy\nTalk (45 min)\n\nModern cloud systems are or
 chestrated by many independent and interacting subsystems\, each specializ
 ing in important services such as scheduling\, data processing and storage
 \, resource management\, etc. Hence\, overall cloud system reliability is 
 determined not only by the reliability of each individual subsystem\, but 
 also by the interactions between them. With recent practices of microservi
 ce and serverless architectures\, each individual subsystem becomes simple
 r and smaller\, while their interactions grow in complexity and diversity.
  We observe that many recent production incidents of large-scale cloud sys
 tems are manifested through failures of the interactions across system bou
 ndaries\, which we term "cross-system interaction failures"\, or "CSI fail
 ures". However\, understanding and addressin
END:VEVENT
BEGIN:VEVENT
UID:af6f15f9-4065-4b26-b327-fa11eecd59a7@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T110000Z
DTEND:20261110T114500Z
SUMMARY:Storytelling as an Incident Management Skill
SEQUENCE:1
LOCATION:Studio
DESCRIPTION:Yashodha Kaur\nTalk (45 min)\n\nManaging incidents well require
 s a number of skills — debugging and systems understanding\, strong comm
 unication\, and high-speed project management — but we rarely talk about
  the power of storytelling in our incident management loop. From oncall pr
 eparation\, to incident handling\, to postmortem creation\, skill at story
 telling can support and even improve our engineering skills. Establishing 
 a setting\, building a coherent narrative\, and understanding your charact
 ers builds memorable and compelling communication that can level up all st
 ages of your incident management. Laura de Vesine is a 20+ year software i
 ndustry veteran. She has spent the last 8 years in SRE working in incident
  analysis and prevention\, chaos engineering\, and the intersection of tec
 hnology and organizational culture. Laura 
END:VEVENT
BEGIN:VEVENT
UID:9423061f-70ae-4b80-8b6b-2b96fd9e232a@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T110000Z
DTEND:20261110T114500Z
SUMMARY:OIDC and CICD: Why Your CI Pipeline Is Your Greatest Security Threa
 t
SEQUENCE:1
LOCATION:Workshop room
DESCRIPTION:Noemí Marrero\nPractice · Lightning talk (10 min)\n\nYour CI/
 CD Process is chock full of credentials\, and almost anyone in your compan
 y has access to it. Configuring your CI correctly is vital to supply chain
  security. We discuss how to reduce that attack surface by enforcing prope
 r branch permissions and using OIDC to reduce long-lived credentials and t
 ie branches to roles.
END:VEVENT
BEGIN:VEVENT
UID:block-lunch-20261110T120000Z@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T120000Z
DTEND:20261110T130000Z
SUMMARY:Lunch
SEQUENCE:0
END:VEVENT
BEGIN:VEVENT
UID:f82b51ac-8a42-40a3-aac2-14e8b2ce13f7@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T133000Z
DTEND:20261110T141500Z
SUMMARY:Synthesizing Sanity with\, and in Spite of\, Synthetic Monitoring
SEQUENCE:1
LOCATION:Main hall
DESCRIPTION:Madeleine Skotnes\nResearch · Workshop (90 min)\n\nSynthetic m
 onitoring\, particularly browser-based monitoring\, is hard to do well. Wh
 en tests pass\, synthetic monitoring provides a uniquely intuitive kind of
  psychological safety - human-like\, verified confidence\, compared to oth
 er forms of monitoring. When tests fail\, synthetic monitoring is often bl
 amed as flaky\, misconfigured\, or unreliable. If not properly implemented
 \, it can not only be financially\, mentally and organisationally draining
 \, but damaging to real customer experience. This talk is a conceptual and
  technical story of 4 years working with Atlassian’s in-house synthetic 
 monitoring solution\, being the owning developer for a tool actively used 
 by 30-40 internal teams to build and manage synthetic monitoring for Jira.
  How can we make synthetic monitoring better serve its purpose 
END:VEVENT
BEGIN:VEVENT
UID:51335477-1f03-46d1-a0b6-bef489442af3@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T133000Z
DTEND:20261110T141500Z
SUMMARY:Measuring Reliability Culture to Optimize Tradeoffs: Perspectives f
 rom an Anthropologist
SEQUENCE:1
LOCATION:Studio
DESCRIPTION:Carolina Concepción\nPractice · Poster / ePoster\n\nUnderstan
 ding how culture influences engineering practices is often a black box. At
  Meta\, we focused on transforming our mantra of “move fast and break th
 ings\,” to “move fast with stable infrastructure” by giving attentio
 n to the cultural elements of doing reliability work. This talk describes 
 this process and decodes how to systematically measure the on-the-ground p
 erspectives of engaging with reliability work so that we can have an infor
 med perspective on how to best optimize the right degree of reliability. T
 he audience will take away actionable practices that can be used to unders
 tand how to evaluate their underlying reliability culture\, take data-driv
 en approaches to measuring reliability sentiment and barriers and facilita
 tors to performing the work\, and identify practices that allow
END:VEVENT
BEGIN:VEVENT
UID:596bf8a7-5778-48d9-b4dd-fea216444d95@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261110T133000Z
DTEND:20261110T141500Z
SUMMARY:The Sins of High Cardinality
SEQUENCE:1
LOCATION:Workshop room
DESCRIPTION:Milja Joki\nSystems · Poster / ePoster\n\nHigh cardinality is 
 a sin in observability metrics collection. Cardinality in the form of gran
 ular labels or other metadata can cause exponential growth in your observa
 bility time-series storage and compute resources\; costing money and slowi
 ng down queries. It's not always economical nor practical to collect all p
 ossible metrics with all possible labels and then worry about how to extra
 ct value using queries after the fact. Sure\, we want as much granularity 
 as possible in our observability data\, but it's a trade-off\, we need to 
 be strategic in using metric cardinality to get the granularity needed to 
 discover and remediate problems. This talk will focus on presenting differ
 ent strategies to constrain the impact of high metrics cardinality referen
 cing applicable open source Prometheus metri
END:VEVENT
BEGIN:VEVENT
UID:a5518cf6-9d15-457a-8853-5d3ed54a39f6@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261111T090000Z
DTEND:20261111T094500Z
SUMMARY:Scam or Savings? A Cloud vs. On-Prem Economic Slapfight
SEQUENCE:1
LOCATION:Main hall
DESCRIPTION:Harry Lee\nOps · Talk (25 min)\n\nSince not giving a single cr
 ap about money turned out to be a purely zero-interest-rate phenomenon\, t
 he "cloud is overpriced vs. no it is not" argument has reignited—usually
  championed on either side by somebody with something to sell you. The spe
 aker has nothing to sell you\, and many truths to tell. It's time to find 
 out where the truth is hiding.
END:VEVENT
BEGIN:VEVENT
UID:aee40108-0a8e-46c0-a1cb-677e896f5c29@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261111T090000Z
DTEND:20261111T094500Z
SUMMARY:Is It Already Time To Version Observability? (Signs Point To Yes.)
SEQUENCE:1
LOCATION:Studio
DESCRIPTION:Aline Yildirim\nOps · Lightning talk (10 min)\n\nPillars\, car
 dinality\, metrics\, dashboards ... the definition of observability has be
 en debated to death\, and I'm done with it. Let's just say that observabil
 ity is a property of complex systems\, just like reliability or performanc
 e. This definition feels both useful and true\, and I am 100% behind it. H
 owever\, there has recently been a generational sea change in data types\,
  usability\, workflows\, and cost models\, along with what users report is
  a massive\, discontinuous leap in value. In the parlance of semantic vers
 ioning\, it is a breaking\, backwards-incompatible change. Which means it
 ’s time to bump the major version number. Observability 1.0\, meet Obser
 vability 2.0. In this presentation\, we will outline the technical and soc
 iotechnical characteristics of each generation of tooling and desc
END:VEVENT
BEGIN:VEVENT
UID:13266087-c842-4bc3-a63e-7402d7a321fe@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261111T090000Z
DTEND:20261111T094500Z
SUMMARY:Automating Disaster Recovery: The Ultimate Reliability Challenge
SEQUENCE:1
LOCATION:Workshop room
DESCRIPTION:Jazzy Van Olderen\nOps · Workshop (90 min)\n\nHere's how I exp
 lain my job to non-techies: if a meteor struck our servers\, it's on my te
 am to fix it. But what if it did? Realistically\, what would happen if a m
 eteor struck your datacenter? Here's the story of a vision\, one to fully 
 automate disaster recovery away\, how I pushed back on it claiming it was 
 impossible\, and how we still executed on it to great success.
END:VEVENT
BEGIN:VEVENT
UID:88194fba-89c2-495b-938b-168ffbf8dbe4@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261111T100000Z
DTEND:20261111T104500Z
SUMMARY:Compliance & Regulatory Standards Are NOT Incompatible with Modern 
 Development Best\n## Practices
SEQUENCE:1
LOCATION:Main hall
DESCRIPTION:Esther Giménez\nPractice · Talk (45 min)\n\nModern software d
 evelopment is all about fast feedback loops\, with best practices like tes
 ting in production\, continuous delivery\, observability driven developmen
 t\, and feature flags. Yet I often hear people complaining that only start
 ups can get away with doing these things\; real\, grown-up companies are s
 ubject to regulatory oversight\, which prevents engineers from deploying t
 heir own code due to separation of concerns\, requires managers to sign of
 f on changes\, etc.
END:VEVENT
BEGIN:VEVENT
UID:33924869-6ceb-49ee-9e23-12d53d6a7bfd@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261111T100000Z
DTEND:20261111T104500Z
SUMMARY:What Can You See from Here?
SEQUENCE:1
LOCATION:Studio
DESCRIPTION:Marcus Mccoy\nOps · Poster / ePoster\n\nWhat you can see depen
 ds on where you stand. Your vantage point plays a big part in how you work
 \, what you think is important\, and how you interpret what's going on aro
 und you. Even your job title or team name can influence–and limit!–how
  you see the world. It's easy to find yourself stuck: no longer learning\,
  polishing a service nobody else cares about\, or wondering why it's so ha
 rd to get anything done outside your own team. A change in perspective can
  change what you think is important\, how you influence the decisions that
  you care about\, and even what you think is possible for yourself. In thi
 s talk\, we'll look at how to get a broader view.
END:VEVENT
BEGIN:VEVENT
UID:block-coffee-and-posters-20261111T110000Z@saas-killa
DTSTAMP:20260820T011621Z
DTSTART:20261111T110000Z
DTEND:20261111T113000Z
SUMMARY:Coffee and posters
SEQUENCE:0
END:VEVENT
END:VCALENDAR