Recherche avancée

Sur d’autres sites (6818)

• Ffmpeg HLS conversion error on AWS Lambda - ffmpeg was killed with signal SIGSEGV

  20 octobre 2023, par RtiM0

  I'm trying to convert video files into HLS streams on a AWS Lambda. The ffmpeg configuration I have setup works for normal (Non HLS) transcoding, but in case of HLS it throws the following error :

  


  stderr:
frame=  341 fps= 84 q=34.0 q=32.0 q=28.0 size=N/A time=00:00:12.52 bitrate=N/A speed= 3.1x    
frame=  385 fps= 85 q=34.0 q=31.0 q=27.0 size=N/A time=00:00:13.97 bitrate=N/A speed=3.08x    
frame=  433 fps= 86 q=33.0 q=30.0 q=27.0 size=N/A time=00:00:15.55 bitrate=N/A speed=3.08x    
[hls @ 0x702c480] Cannot use rename on non file protocol, this may lead to races and temporary partial files
[hls @ 0x702c480] Opening '/tmp/stream_0.m3u8' for writing
[hls @ 0x702c480] Opening '/tmp/stream_1.m3u8' for writing
[hls @ 0x702c480] Opening '/tmp/stream_2.m3u8' for writing
[hls @ 0x702c480] Opening '/tmp/master.m3u8' for writing
ffmpeg was killed with signal SIGSEGV


  


  And I error thrown by fluent-ffmpeg is this :

  


  2023-03-06T10:21:44.555Z    15a0a43b-5e24-42ac-ae72-fe04f0c72a3e    ERROR   Invoke Error    &#xA;{&#xA;    "errorType": "Error",&#xA;    "errorMessage": "ffmpeg was killed with signal SIGSEGV",&#xA;    "stack": [&#xA;        "Error: ffmpeg was killed with signal SIGSEGV",&#xA;        "    at ChildProcess.<anonymous> (/var/task/node_modules/fluent-ffmpeg/lib/processor.js:180:22)",&#xA;        "    at ChildProcess.emit (node:events:513:28)",&#xA;        "    at ChildProcess._handle.onexit (node:internal/child_process:291:12)"&#xA;    ]&#xA;}&#xA;</anonymous>

  


  This is the code I run on AWS Lambda to convert video files into HLS :

  


  export const compressToHLS = (sourcePath, outputFolder) =>
  new Promise((resolve, reject) => {
    Ffmpeg(sourcePath)
      .complexFilter([
        {
          filter: "split",
          options: "3",
          inputs: "v:0",
          outputs: ["v1", "v2", "v3"],
        },
        {
          filter: "scale",
          options: {
            w: 1280,
            h: 720,
          },
          inputs: "v1",
          outputs: "v1out",
        },
        {
          filter: "scale",
          options: {
            w: 960,
            h: 540,
          },
          inputs: "v2",
          outputs: "v2out",
        },
        {
          filter: "scale",
          options: {
            w: 640,
            h: 360,
          },
          inputs: "v3",
          outputs: "v3out",
        },
      ])
      .outputOptions([
        "-map [v1out]",
        "-c:v:0",
        "libx264",
        "-b:v 3000000",
        "-map [v2out]",
        "-c:v:1",
        "libx264",
        "-b:v 2000000",
        "-map [v3out]",
        "-c:v:2",
        "libx264",
        "-b:v 1000000",
      ])
      .outputOptions([
        "-map a:0",
        "-c:a:0 aac",
        "-b:a:0 96000",
        "-ar 48000",
        "-ac 2",
        "-map a:0",
        "-c:a:1 aac",
        "-b:a:1 96000",
        "-ar 48000",
        "-ac 2",
        "-map a:0",
        "-c:a:2 aac",
        "-b:a:2 96000",
        "-ar 48000",
        "-ac 2",
      ])
      .outputOptions([
        "-f hls",
        "-hls_time 10",
        "-hls_playlist_type vod",
        "-hls_flags independent_segments",
        "-hls_segment_type mpegts",
        `-hls_segment_filename ${outputFolder}/%v_%d.ts`,
        "-master_pl_name master.m3u8",
      ])
      .outputOption("-var_stream_map", "v:0,a:0 v:1,a:1 v:2,a:2")
      .outputOption("-preset veryfast")
      .output(`${outputFolder}/stream_%v.m3u8`)
      .on("start", (cmdline) => console.log(cmdline))
      .on("progress", (progress) => {
        let prog = Math.floor(progress.percent * 10) / 10;
        if (Math.round(prog) % 10 == 0) {
          console.log(`${prog}% complete`);
        }
      })
      .on("error", (err, stdout, stderr) => {
        if (err) {
          console.log(err.message);
          console.log("stdout:\n" + stdout);
          console.log("stderr:\n" + stderr);
          reject(err);
        }
      })
      .on("end", () => resolve())
      .run();
  });


  


  In this code the sourcePath is usually a presigned URL from S3 (But I have also tried to download a file on /tmp and setting sourcePath as the path to the downloaded file) and outputFolder is tmpdir() which is /tmp.

  


  I have the lambda settings configured to have 10GB of memory and 10GB of ephemeral storage (the maximum allowed).

  


• ffmpeg is failing to load shared libraries after a ./configure with a prefix inside a conda environment

  30 janvier 2024, par user3133806

  I am using conda and building ffmpeg from source within that environment.

  


  I ran the following commands :

  


  conda create --name my_conda_env
conda activate my_conda_env
# Now I am in the conda environment
# $CONDA_PREFIX is /home/myuser/.conda/envs/my_conda_env/bin/ffmpeg

# Checkout ffmpeg code
# git checkout ...

./configure --prefix=$CONDA_PREFIX --enable-shared --disable-static && make distclean && make -j 100 && make install

# The above command does install the newly built ffmpeg into:
# /home/myuser/.conda/envs/my_conda_env/bin/ffmpeg

# However it fails to execute:
ffmpeg
ffmpeg: error while loading shared libraries: libavdevice.so.58: cannot open shared object file: No such file or directory

# When I add the conda lib path to LD_LIBRARY_PATH it works:
LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH ffmpeg
ffmpeg version n4.2.9-4-gd7beb0c61f Copyright (c) 2000-2023 the FFmpeg developers

# I thought ./configure with a --prefix will build a binary that will search for libraries relative to itself, but that does not appear to be the case:

strace -o /tmp/strace.out ffmpeg
tail /tmp/strace.out

openat(AT_FDCWD, "/usr/lib64/haswell/x86_64/libavdevice.so.58", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
newfstatat(AT_FDCWD, "/usr/lib64/haswell/x86_64", 0x7fff65f56d90, 0) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/usr/lib64/haswell/libavdevice.so.58", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
newfstatat(AT_FDCWD, "/usr/lib64/haswell", 0x7fff65f56d90, 0) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/usr/lib64/avx512_1/x86_64/libavdevice.so.58", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
newfstatat(AT_FDCWD, "/usr/lib64/avx512_1/x86_64", 0x7fff65f56d90, 0) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/usr/lib64/avx512_1/libavdevice.so.58", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
newfstatat(AT_FDCWD, "/usr/lib64/avx512_1", 0x7fff65f56d90, 0) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/usr/lib64/x86_64/libavdevice.so.58", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
newfstatat(AT_FDCWD, "/usr/lib64/x86_64", 0x7fff65f56d90, 0) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/usr/lib64/libavdevice.so.58", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
newfstatat(AT_FDCWD, "/usr/lib64", {st_mode=S_IFDIR|0555, st_size=49526, ...}, 0) = 0
writev(2, [{iov_base="ffmpeg", iov_len=6}, {iov_base=": ", iov_len=2}, {iov_base="error while loading shared libra"..., iov_len=36}, {i
ov_base=": ", iov_len=2}, {iov_base="libavdevice.so.58", iov_len=17}, {iov_base=": ", iov_len=2}, {iov_base="cannot open shared object file", iov_len=30}, {iov_base=": ", iov_len=2}, {iov_base="No such file or directory", iov_len=25}, {iov_base="\n", iov_len=1}], 10) = 
123


  


  Conda documentation says not to use LD_LIBRARY_PATH here :

  


  https://docs.conda.io/projects/conda-build/en/stable/resources/use-shared-libraries.html#shared-libraries-in-macos-and-linux

  


  How can I build ffmpeg from source in a conda environment and have the binary find the .so file relative to itself ?

  


• What is Behavioural Segmentation and Why is it Important ?

  28 septembre 2023, par Erin — Analytics Tips

  Amidst the dynamic landscape of web analytics, understanding customers has grown increasingly vital for businesses to thrive. While traditional demographic-focused strategies possess merit, they need to uncover the nuanced intricacies of individual online behaviours and preferences. As customer expectations evolve in the digital realm, enterprises must recalibrate their approaches to remain relevant and cultivate enduring digital relationships.

  In this context, the surge of technology and advanced data analysis ushers in a marketing revolution : behavioural segmentation. Businesses can unearth invaluable insights by meticulously scrutinising user actions, preferences and online interactions. These insights lay the foundation for precisely honed, high-performing, personalised campaigns. The era dominated by blanket, catch-all marketing strategies is yielding to an era of surgical precision and tailored engagement. 

  While the insights from user behaviours empower businesses to optimise customer experiences, it’s essential to strike a delicate balance between personalisation and respecting user privacy. Ethical use of behavioural data ensures that the power of segmentation is wielded responsibly and in compliance, safeguarding user trust while enabling businesses to thrive in the digital age.

  What is behavioural segmentation ?

  Behavioural segmentation is a crucial concept in web analytics and marketing. It involves categorising individuals or groups of users based on their online behaviour, actions and interactions with a website. This segmentation method focuses on understanding how users engage with a website, their preferences and their responses to various stimuli. Behavioural segmentation classifies users into distinct segments based on their online activities, such as the pages they visit, the products they view, the actions they take and the time they spend on a site.

  Behavioural segmentation plays a pivotal role in web analytics for several reasons :

  1. Enhanced personalisation :

  Understanding user behaviour enables businesses to personalise online experiences. This aids with delivering tailored content and recommendations to boost conversion, customer loyalty and customer satisfaction.

  2. Improved user experience :

  Behavioural segmentation optimises user interfaces (UI) and navigation by identifying user paths and pain points, enhancing the level of engagement and retention.

  3. Targeted marketing :

  Behavioural segmentation enhances marketing efficiency by tailoring campaigns to user behaviour. This increases the likelihood of interest in specific products or services.

  4. Conversion rate optimisation :

  Analysing behavioural data reveals factors influencing user decisions, enabling website optimisation for a streamlined purchasing process and higher conversion rates.

  5. Data-driven decision-making :

  Behavioural segmentation empowers data-driven decisions. It identifies trends, behavioural patterns and emerging opportunities, facilitating adaptation to changing user preferences and market dynamics.

  6. Ethical considerations :

  Behavioural segmentation provides valuable insights but raises ethical concerns. User data collection and use must prioritise transparency, privacy and responsible handling to protect individuals’ rights.

  The significance of ethical behavioural segmentation will be explored more deeply in a later section, where we will delve into the ethical considerations and best practices for collecting, storing and utilising behavioural data in web analytics. It’s essential to strike a balance between harnessing the power of behavioural segmentation for business benefits and safeguarding user privacy and data rights in the digital age.

  

  Different types of behavioural segments with examples

  1. Visit-based segments : These segments hinge on users’ visit patterns. Analyse visit patterns, compare first-time visitors to returning ones, or compare users landing on specific pages to those landing on others.
     • Example : The real estate website Zillow can analyse how first-time visitors and returning users behave differently. By understanding these patterns, Zillow can customise its website for each group. For example, they can highlight featured listings and provide navigation tips for first-time visitors while offering personalised recommendations and saved search options for returning users. This could enhance user satisfaction and boost the chances of conversion.
  2. Interaction-based segments : Segments can be created based on user interactions like special events or goals completed on the site.
     • Example : Airbnb might use this to understand if users who successfully book accommodations exhibit different behaviours than those who don’t. This insight could guide refinements in the booking process for improved conversion rates.
  3. Campaign-based segments : Beyond tracking visit numbers, delve into usage differences of visitors from specific sources or ad campaigns for deeper insights.
     • Example : Nike might analyse user purchase behaviour from various traffic sources (referral websites, organic, direct, social media and ads). This informs marketing segmentation adjustments, focusing on high-performance channels. It also customises the website experience for different traffic sources, optimising content, promotions and navigation. This data-driven approach could boost user experiences and maximise marketing impact for improved brand engagement and sales conversions.
  4. Ecommerce segments : Separate users based on purchases, even examining the frequency of visits linked to specific products. Segment heavy users versus light users. This helps uncover diverse customer types and browsing behaviours.
     • Example : Amazon could create segments to differentiate between visitors who made purchases and those who didn’t. This segmentation could reveal distinct usage patterns and preferences, aiding Amazon in tailoring its recommendations and product offerings.
  5. Demographic segments : Build segments based on browser language or geographic location, for instance, to comprehend how user attributes influence site interactions.
     • Example : Netflix can create user segments based on demographic factors like geographic location to gain insight into how a visitor’s location can influence content preferences and viewing behaviour. This approach could allow for a more personalised experience.
  6. Technographic segments : Segment users by devices or browsers, revealing variations in site experience and potential platform-specific issues or user attitudes.
     • Example : Google could create segments based on users’ devices (e.g., mobile, desktop) to identify potential issues in rendering its search results. This information could be used to guide Google in providing consistent experiences regardless of device.

  

  The importance of ethical behavioural segmentation

  Respecting user privacy and data protection is crucial. Matomo offers features that align with ethical segmentation practices. These include :

  • Anonymization : Matomo allows for data anonymization, safeguarding individual identities while providing valuable insights.
  • GDPR compliance : Matomo is GDPR compliant, ensuring that user data is handled following European data protection regulations.
  • Data retention and deletion : Matomo enables businesses to set data retention policies and delete user data when it’s no longer needed, reducing the risk of data misuse.
  • Secured data handling : Matomo employs robust security measures to protect user data, reducing the risk of data breaches.

  Real-world examples of ethical behavioural segmentation :

  1. Content publishing : A leading news website could utilise data anonymization tools to ethically monitor user engagement. This approach allows them to optimise content delivery based on reader preferences while ensuring the anonymity and privacy of their target audience.
  2. Non-profit organisations : A charity organisation could embrace granular user control features. This could be used to empower its donors to manage their data preferences, building trust and loyalty among supporters by giving them control over their personal information.

  

  Examples of effective behavioural segmentation

  Companies are constantly using behavioural insights to engage their audiences effectively. In this section, we’ll delve into real-world examples showcasing how top companies use behavioural segmentation to enhance their marketing efforts.

  

  1. Coca-Cola’s behavioural insights for marketing strategy : Coca-Cola employs behavioural segmentation to evaluate its advertising campaigns. Through analysing user engagement across TV commercials, social media promotions and influencer partnerships, Coca-Cola’s marketing team can discover that video ads shared by influencers generate the highest ROI and web traffic.

     This insight guides the reallocation of resources, leading to increased sales and a more effective advertising strategy.

  2. eBay’s custom conversion approach : eBay excels in conversion optimisation through behavioural segmentation. When users abandon carts, eBay’s dynamic system sends personalised email reminders featuring abandoned items and related recommendations tailored to user interests and past purchase decisions.

     This strategy revives sales, elevates conversion rates and sparks engagement. eBay’s adeptness in leveraging behavioural insights transforms user experience, steering a customer journey toward conversion.

  3. Sephora’s data-driven conversion enhancement : Data analysts can use Sephora’s behavioural segmentation strategy to fuel revenue growth through meticulous data analysis. By identifying a dedicated subset of loyal customers who exhibit a consistent preference for premium skincare products, data analysts enable Sephora to customise loyalty programs.

     These personalised rewards programs provide exclusive discounts and early access to luxury skincare releases, resulting in heightened customer engagement and loyalty. The data-driven precision of this approach directly contributes to amplified revenue from this specific customer segment.

  Examples of the do’s and don’ts of behavioural segmentation 

  

  Behavioural segmentation is a powerful marketing and data analysis tool, but its success hinges on ethical and responsible practices. In this section, we will explore real-world examples of the do’s and don’ts of behavioural segmentation, highlighting companies that have excelled in their approach and those that have faced challenges due to lapses in ethical considerations.

  Do’s of behavioural segmentation :

  • Personalised messaging :
    • Example : Spotify
      • Spotify’s success lies in its ability to use behavioural data to curate personalised playlists and user recommendations, enhancing its music streaming experience.
  • Transparency :
    • Example : Basecamp
      • Basecamp’s transparency in sharing how user data is used fosters trust. They openly communicate data practices, ensuring users are informed and comfortable.
  • Anonymization
    • Example : Matomo’s anonymization features
      • Matomo employs anonymization features to protect user identities while providing valuable insights, setting a standard for responsible data handling.
  • Purpose limitation :
    • Example : Proton Mail
      • Proton Mail strictly limits the use of user data to email-related purposes, showcasing the importance of purpose-driven data practices.
  • Dynamic content delivery : 
    • Example : LinkedIn
      • LinkedIn uses behavioural segmentation to dynamically deliver job recommendations, showcasing the potential for relevant content delivery.
  • Data security :
    • Example : Apple
      • Apple’s stringent data security measures protect user information, setting a high bar for safeguarding sensitive data.
  • Adherence to regulatory compliance : 
    • Example : Matomo’s regulatory compliance features
      • Matomo’s regulatory compliance features ensure that businesses using the platform adhere to data protection regulations, further promoting responsible data usage.

  Don’ts of behavioural segmentation :

  • Ignoring changing regulations
    • Example : Equifax
      • Equifax faced major repercussions for neglecting evolving regulations, resulting in a data breach that exposed the sensitive information of millions.
  • Sensitive attributes
    • Example : Twitter
      • Twitter faced criticism for allowing advertisers to target users based on sensitive attributes, sparking concerns about user privacy and data ethics.
  • Data sharing without consent
    • Example : Meta & Cambridge Analytica
      • The Cambridge Analytica scandal involving Meta (formerly Facebook) revealed the consequences of sharing user data without clear consent, leading to a breach of trust.
  • Lack of control
    • Example : Uber
      • Uber faced backlash for its poor data security practices and a lack of control over user data, resulting in a data breach and compromised user information.
  • Don’t be creepy with invasive personalisation
    • Example : Offer Moment
      • Offer Moment’s overly invasive personalisation tactics crossed ethical boundaries, unsettling users and eroding trust.

  These examples are valuable lessons, emphasising the importance of ethical and responsible behavioural segmentation practices to maintain user trust and regulatory compliance in an increasingly data-driven world.

  Continue the conversation

  Diving into customer behaviours, preferences and interactions empowers businesses to forge meaningful connections with their target audience through targeted marketing segmentation strategies. This approach drives growth and fosters exceptional customer experiences, as evident from the various common examples spanning diverse industries.

  In the realm of ethical behavioural segmentation and regulatory compliance, Matomo is a trusted partner. Committed to safeguarding user privacy and data integrity, our advanced web analytics solution empowers your business to harness the power of behavioral segmentation, all while upholding the highest standards of compliance with stringent privacy regulations.

  To gain deeper insight into your visitors and execute impactful marketing campaigns, explore how Matomo can elevate your efforts. Try Matomo free for 21-days, no credit card required.