{"id":2900,"date":"2026-05-30T14:25:37","date_gmt":"2026-05-30T14:25:37","guid":{"rendered":"https:\/\/www.serverspan.com\/blog\/"},"modified":"2026-05-30T14:25:38","modified_gmt":"2026-05-30T14:25:38","slug":"containerul-docker-mi-a-mancat-tot-ram-ul-de-pe-vps-cum-diagnostichezi-si-repari-oom-kills","status":"publish","type":"post","link":"https:\/\/www.serverspan.ro\/blog\/containerul-docker-mi-a-mancat-tot-ram-ul-de-pe-vps-cum-diagnostichezi-si-repari-oom-kills\/","title":{"rendered":"Containerul Docker mi-a m\u00e2ncat tot RAM-ul de pe VPS: cum diagnostichezi \u0219i repari OOM Kills"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">Containerul t\u0103u a rulat brici s\u0103pt\u0103m\u00e2ni la r\u00e2nd. Apoi \u00eencepe brusc s\u0103 se restarteze la fiecare c\u00e2teva ore. Te ui\u021bi \u00een loguri \u0219i e bezn\u0103 total\u0103, nicio eroare util\u0103. Dai un <code>docker ps<\/code> \u0219i vezi c\u0103 statusul face ping-pong \u00eentre <code>Up<\/code> \u0219i <code>Restarting<\/code>. Singurul indiciu real este un exit code 137, care \u00eenseamn\u0103 c\u0103 procesul a primit un SIGKILL direct. Pe un VPS, \u00een 99% din cazuri, asta e semn\u0103tura clar\u0103 c\u0103 Linux Out-Of-Memory (OOM) killer-ul a intrat \u00een ac\u021biune.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Acest articol \u00ee\u021bi arat\u0103 exact workflow-ul de debugging pentru a confirma un OOM kill, a depista procesul care a umplut memoria \u0219i a rezolva cauza de fond. F\u0103r\u0103 dat cu presupusul. F\u0103r\u0103 restarturi disperate ale VPS-ului \u00een speran\u021ba c\u0103 problema \u201ese rezolv\u0103 de la sine\u201d.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Ce \u00eenseamn\u0103 de fapt exit code 137<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Exit code 137 este rezultatul matematic a dou\u0103 numere: 128 (baza pentru procesele terminate printr-un semnal extern) plus 9 (semnalul SIGKILL). C\u00e2nd Docker \u00ee\u021bi arunc\u0103 eroarea asta, \u00eenseamn\u0103 c\u0103 procesul principal din container a fost oprit for\u021bat, f\u0103r\u0103 drept de apel, de o entitate extern\u0103.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Exist\u0103 doar dou\u0103 explica\u021bii logice:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>OOM killer-ul din kernel-ul Linux<\/strong> a omor\u00e2t procesul pentru c\u0103 sistemul a r\u0103mas f\u0103r\u0103 RAM \u0219i avea nevoie disperat\u0103 s\u0103 elibereze memorie ca s\u0103 nu crape OS-ul.<\/li>\n\n\n\n<li><strong>Cineva (sau un script de automatizare) a executat manual<\/strong> <code>docker kill<\/code> sau <code>kill -9<\/code> pe acel PID.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Dac\u0103 nu i-ai dat tu kill inten\u021bionat, e clar OOM killer-ul. Aten\u021bie: OOM killer-ul este un mecanism vital al kernel-ului Linux, nu o inven\u021bie Docker. E prezent pe orice distribu\u021bie \u0219i se activeaz\u0103 automat c\u00e2nd kernel-ul nu mai are de unde s\u0103 aloce memorie. Rolul lui e cinic, dar necesar: alege un proces \u0219i \u00eel execut\u0103 ca s\u0103 elibereze RAM.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Pe un VPS mic care nu are swap configurat (sau care are deja swap-ul plin ochi), OOM killer-ul nu st\u0103 la discu\u021bii. Love\u0219te fie cel mai mare consumator de RAM, fie ultimul proces pornit, fie unul cu un scor mare de \u201ebadness\u201d calculat de kernel. Victima poate fi baza ta de date, reverse proxy-ul sau chiar daemon-ul Docker \u00een sine.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Pasul 1: Confirm\u0103 c\u0103 a fost OOM kill<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Nu trage concluzii pripite. Bate urm\u0103toarele trei comenzi, fix \u00een ordinea asta, ca s\u0103 fii sigur.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">1. Verific\u0103 docker inspect<\/h3>\n\n\n\n<pre class=\"wp-block-preformatted\">docker inspect NUME_CONTAINER --format='{{.State.OOMKilled}} {{.State.ExitCode}}'<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Dac\u0103 \u00ee\u021bi returneaz\u0103 <code>true 137<\/code>, Docker \u00ee\u021bi confirm\u0103 negru pe alb c\u0103 OOM killer-ul este responsabil. Dac\u0103 scrie <code>false 137<\/code>, \u00eenseamn\u0103 c\u0103 altceva a trimis SIGKILL-ul. Verific\u0103-\u021bi scripturile de monitorizare, pipeline-urile CI\/CD sau istoricul de comenzi.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">2. Verific\u0103 log-urile kernel-ului<\/h3>\n\n\n\n<pre class=\"wp-block-preformatted\">sudo dmesg | grep -i \"killed process\"<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Kernel-ul noteaz\u0103 absolut fiecare ac\u021biune a OOM killer-ului. Ar trebui s\u0103 vezi o linie care arat\u0103 cam a\u0219a:<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">[123456.789012] Out of memory: Killed process 12345 (postgres) total-vm:1048576kB, anon-rss:512000kB, file-rss:0kB, shmem-rss:0kB<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Linia asta e aur: \u00ee\u021bi spune exact ce proces a picat, PID-ul lui \u0219i c\u00e2t\u0103 memorie real\u0103 folosea \u00een acel moment. Dac\u0103 vezi zeci de astfel de linii cu PID-uri diferite, \u00eenseamn\u0103 c\u0103 OOM killer-ul ruleaz\u0103 la foc automat \u0219i VPS-ul t\u0103u e g\u00e2tuit de memorie cronic.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">3. Verific\u0103 docker stats<\/h3>\n\n\n\n<pre class=\"wp-block-preformatted\">docker stats --no-stream<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Comanda asta \u00ee\u021bi arat\u0103 c\u00e2t RAM \u0219i CPU suge fiecare container \u00een timp real. Uit\u0103-te dup\u0103 containere cu un procent mare la <code>MEM %<\/code>. Mai important: dac\u0103 la coloana <code>MEM LIMIT<\/code> vezi valoarea total\u0103 a RAM-ului de pe server la fiecare container \u00een parte, \u00eenseamn\u0103 c\u0103 nu ai setat limite de memorie absolut deloc. Fix aia e cauza de fond.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Pasul 2: G\u0103se\u0219te vinovatul care hale\u0219te memoria<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Odat\u0103 ce ai confirmat OOM kill-ul, trebuie s\u0103 afli ce anume din interiorul containerului a sec\u0103tuit resursele. Iat\u0103 suspec\u021bii de serviciu pentru setup-urile self-hosted pe VPS-uri.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Containere cu modele AI (Machine Learning)<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Orice container care proceseaz\u0103 inferen\u021b\u0103 AI va arunca \u00eentregul model direct \u00een RAM la pornire. Ollama, modulul de machine learning din Immich \u0219i alte servicii similare au nevoie de 2-8 GB de RAM doar ca s\u0103 respire. Dac\u0103 tu ai un VPS de 2 GB, containerul o s\u0103-\u0219i ia OOM kill fix \u00een momentul ini\u021bializ\u0103rii. Nu degeaba specific\u0103m clar \u00een <a href=\"https:\/\/www.serverspan.ro\/blog\/cum-sa-gazduiesti-singur-backup-foto-si-video-immich-pe-vps-serverspan-alternativa-zero-cloud-la-google-photos-ghid-2026\/\">ghidul de self-hosting pentru Immich<\/a> c\u0103 stack-ul de ML cere minimum 2-4 GB de RAM dedica\u021bi exclusiv pentru recunoa\u0219tere facial\u0103.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Baze de date l\u0103sate cu set\u0103rile default la conexiuni<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">PostgreSQL, MariaDB \u0219i Redis aloc\u0103 memorie per conexiune activ\u0103. O configura\u021bie default de Postgres e g\u00e2ndit\u0103 s\u0103 accepte sute de conexiuni, fiecare ciupind c\u00e2\u021biva megabytes de RAM. Matematica e simpl\u0103: dac\u0103 aplica\u021bia ta deschide un connection pool de 100 de conexiuni pe un VPS am\u0103r\u00e2t de 2 GB, baza de date \u00ee\u021bi va culca serverul garantat.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Aplica\u021bii Java f\u0103r\u0103 heap explicit<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Prin defini\u021bie, JVM-ul (Java) \u00ee\u0219i aloc\u0103 automat un heap de 25% din RAM-ul total al sistemului. Pe un server de 4 GB, asta \u00eenseamn\u0103 1 GB luat ostatic doar pentru heap, plus memoria necesar\u0103 pentru thread-uri \u0219i spa\u021biul off-heap. Dac\u0103 ai mai multe servicii Java pornite simultan, ele vor cere \u00eempreun\u0103 mult mai mult RAM dec\u00e2t are host-ul fizic disponibil.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Build-uri rulate live cu <code>docker compose up --build<\/code><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">C\u00e2nd construie\u0219ti imagini Docker direct pe serverul de produc\u021bie (\u00een special proiecte Node.js sau Python), RAM-ul bubuie temporar la maxim. Un simplu <code>npm install<\/code> sau o compilare de pachete poate trigera OOM killer-ul care va ucide alte containere absolut nevinovate din sistem. Este exact avertismentul pe care \u00eel d\u0103m \u00een <a href=\"https:\/\/www.serverspan.ro\/blog\/scapa-de-costurile-vercel-ruleaza-coolify-pe-propriul-vps-si-lanseaza-aplicatii-ca-un-paas\/\">ghidul pentru Coolify<\/a>: un proces greu de build Nixpacks pe un VPS de 2 GB o s\u0103-\u0219i ia OOM kill \u00eenainte s\u0103 apuce s\u0103 termine de desc\u0103rcat dependen\u021bele.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Pipeline-uri brutale de OCR \u0219i documente<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Motoarele OCR precum Tesseract (folosit de Paperless-ngx) \u00eencarc\u0103 documentele pagin\u0103 cu pagin\u0103 \u00een memorie. Dac\u0103 \u00eei arunci un PDF scanat de 100 de pagini, RAM-ul va face un spike uria\u0219 care dureaz\u0103 minute \u00eentregi. \u00cen <a href=\"https:\/\/www.serverspan.ro\/blog\/paperless-ngx-2-20-pe-un-vps-cum-alegi-corect-ocr-ul-ingestia-prin-email-si-layout-ul-de-stocare-din-prima-zi\/\">ghidul de arhitectur\u0103 pentru Paperless-ngx<\/a> am explicat de ce nu po\u021bi rula stack-ul cu mai pu\u021bin de 2 GB RAM, chiar \u0219i pentru un volum mic de documente.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Pasul 3: Cum opre\u0219ti hemoragia acum (Solu\u021bii imediate)<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Ai trei moduri de a opri ciclul de OOM kills, de la bandajul rapid la rezolvarea de fond.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Solu\u021bia A: Pune limite hard de memorie pe container<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Limitezi direct &#8220;blast radius-ul&#8221;. Dac\u0103 limitezi RAM-ul la nivel de container \u0219i acesta face overcommit, Docker va ucide strict acel container. Restul sistemului, baza de date \u0219i celelalte servicii r\u0103m\u00e2n neatinse. Adaug\u0103 blocul \u0103sta \u00een <code>docker-compose.yml<\/code>:<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">services:\n  app:\n    deploy:\n      resources:\n        limits:\n          memory: 512M\n          cpus: \"1.0\"<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Ajusteaz\u0103 valoarea pe logica serviciului. Un Postgres mic poate sta \u00een 1 GB. Un API simplu de Node.js poate fi str\u00e2ns la 256 MB. Regula de aur e s\u0103 pui o limit\u0103 global\u0103 adunat\u0103 care s\u0103 fie sub memoria fizic\u0103 a serverului t\u0103u.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Solu\u021bia B: Configureaz\u0103 un fi\u0219ier de Swap<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Swap-ul nu \u00eenlocuie\u0219te RAM-ul, dar \u00ee\u021bi cump\u0103r\u0103 timp. C\u00e2nd memoria se umple, kernel-ul mut\u0103 datele inactive pe SSD. Asta te salveaz\u0103 de la OOM kills \u00een timpul spike-urilor temporare (cum ar fi un build Docker sau importul unui fi\u0219ier imens).<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">sudo fallocate -l 2G \/swapfile\nsudo chmod 600 \/swapfile\nsudo mkswap \/swapfile\nsudo swapon \/swapfile<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Pentru persisten\u021b\u0103 dup\u0103 reboot, lipe\u0219te asta \u00een <code>\/etc\/fstab<\/code>:<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">\/swapfile none swap sw 0 0<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Pe un VPS cu SSD, 2 GB de swap sunt de ajuns pentru absorb\u021bia \u0219ocurilor. Nu pica \u00een capcana de a te baza pe swap pentru consum sus\u021binut. Dac\u0103 containerul t\u0103u face swap \u00eencontinuu, \u00ee\u021bi omoar\u0103 IOPS-urile discului; are nevoie de RAM real, nu de bandaje.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Solu\u021bia C: F\u0103 upgrade la VPS (Scale up)<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Dac\u0103 ai pus deja limite, ai ad\u0103ugat swap, dar containerele \u00eenc\u0103 se lovesc de tavan, \u00eenseamn\u0103 c\u0103 e\u0219ti subdimensionat. E o poveste clasic\u0103 \u00een self-hosting: \u00eencepi cu un simplu reverse proxy, pui un wiki, arunci \u0219i o baz\u0103 de date, iar 2 GB de RAM ajung s\u0103 fie o glum\u0103 proast\u0103.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">La noi, po\u021bi face upgrade la un <a href=\"https:\/\/www.serverspan.ro\/virtual-servers\">VPS KVM ServerSpan<\/a> on-the-fly, f\u0103r\u0103 s\u0103 pierzi date. Dac\u0103 treci la 4 GB RAM, ai loc suficient pentru a respira f\u0103r\u0103 s\u0103 fii for\u021bat s\u0103 faci micro-management la swap \u0219i memory limits \u00een fiecare s\u0103pt\u0103m\u00e2n\u0103.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Pasul 4: F\u0103 tuning direct \u00een interiorul aplica\u021biei<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Limitele din Docker previn crash-ul \u00eentregului OS. Ca s\u0103 previi crash-ul containerului, trebuie s\u0103-i tai pofta de RAM direct din fi\u0219ierele lui de configurare.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">PostgreSQL: str\u00e2nge cureaua la max_connections<\/h3>\n\n\n\n<pre class=\"wp-block-preformatted\">max_connections = 50\nshared_buffers = 128MB<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Nu l\u0103sa pe default (100+ conexiuni) pe un server mic. 50 de conexiuni pentru un setup de self-hosting pe un VPS de 2 GB sunt mai mult dec\u00e2t suficiente.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Redis: for\u021beaz\u0103 maxmemory<\/h3>\n\n\n\n<pre class=\"wp-block-preformatted\">maxmemory 256mb\nmaxmemory-policy allkeys-lru<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Dac\u0103 nu-l limitezi, Redis o s\u0103 tot scrie \u00een RAM p\u00e2n\u0103 d\u0103 de OOM killer. Setarea <code>allkeys-lru<\/code> cur\u0103\u021b\u0103 elegant cheile vechi \u0219i nefolosite atunci c\u00e2nd atinge cei 256 MB.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Java: seteaz\u0103 clar Xmx \u0219i Xms<\/h3>\n\n\n\n<pre class=\"wp-block-preformatted\">-Xmx512m -Xms256m<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Opre\u0219te JVM-ul din a-\u0219i aloca aberant un sfert din RAM-ul VPS-ului t\u0103u. A\u0219a \u00eel blochezi fix la 512 MB de heap.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Nextcloud: limiteaz\u0103 workerii PHP-FPM<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Imaginile oficiale de Nextcloud vin cu pool-uri agresive pe PHP-FPM. Dac\u0103 la\u0219i 50 de workeri pe 2 GB de RAM, \u00ee\u021bi va cr\u0103pa la prima sincronizare serioas\u0103 a desktop client-ului. Vezi \u00een <a href=\"https:\/\/www.serverspan.ro\/blog\/nextcloud-30-pe-vps-da-timeout-repara-php-fpm-redis-cron-mariadb-si-reverse-proxy-ul-ca-instanta-sa-nu-moara-dupa-3-luni\/\">ghidul nostru de stabilitate Nextcloud<\/a> formula matematic\u0103 clar\u0103 ca s\u0103 setezi fix num\u0103rul de workeri pe care VPS-ul t\u0103u \u00eei poate duce.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Pasul 5: Pune monitorizare ca s\u0103 vezi dezastrul \u00eenainte s\u0103 crape<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">OOM kill-ul n-ar trebui s\u0103 fie o surpriz\u0103. P\u0103ze\u0219te-\u021bi spatele cu ni\u0219te alerte de baz\u0103.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Urm\u0103re\u0219te consumul live din terminal<\/h3>\n\n\n\n<pre class=\"wp-block-preformatted\">watch -n 5 docker stats --no-stream --format \"table {{.Name}}\\t{{.MemUsage}}\\t{{.MemPerc}}\"<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Comanda face refresh la fiecare 5 secunde. Dac\u0103 vezi un container la care procentul cre\u0219te continuu, ai un memory leak (scurgere de memorie).<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Verific\u0103 corect RAM-ul cu comanda free<\/h3>\n\n\n\n<pre class=\"wp-block-preformatted\">free -h<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Regula num\u0103rul 1 la sysadmini pe Linux: nu te uita la coloana <code>free<\/code>, uit\u0103-te la <code>available<\/code>. Linux-ul bag\u0103 RAM-ul \u201eliber\u201d \u00een buffere de disc, a\u0219a c\u0103 valoarea <code>available<\/code> \u00ee\u021bi arat\u0103 c\u00e2t RAM este efectiv disponibil pentru o alocare nou\u0103. Dac\u0103 scade sub 10% din total, urmeaz\u0103 un OOM kill.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Bag\u0103 un script de alertare \u00een cron<\/h3>\n\n\n\n<pre class=\"wp-block-preformatted\">#!\/bin\/bash\nAVAILABLE=$(free | awk '\/Mem:\/ {printf \"%.0f\", $7\/$2 * 100}')\nif [ \"$AVAILABLE\" -lt 15 ]; then\n  echo \"AVERTIZARE: Doar ${AVAILABLE}% RAM disponibil pe $(hostname)\" | logger\nfi<\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Pune asta s\u0103 ruleze din minut \u00een minut. C\u00e2nd <code>available<\/code> scade sub 15%, scrie o alert\u0103 \u00een syslog. De aici, po\u021bi trimite simplu log-ul pe email sau spre un webhook de Discord\/Telegram.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Workflow-ul clar de troubleshooting pentru OOM<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Nu s\u0103ri etapele. Urmeaz\u0103 exact pa\u0219ii \u0103\u0219tia data viitoare c\u00e2nd moare un container:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Dai un <code>docker inspect<\/code>. E <code>OOMKilled<\/code> pe true? Dac\u0103 nu, ia la puricat scripturile c\u0103 cineva i-a dat kill manual.<\/li>\n\n\n\n<li>Dai un <code>dmesg | grep \"killed process\"<\/code>. Afli cine a picat \u0219i cu c\u00e2t RAM \u00een bra\u021be.<\/li>\n\n\n\n<li>Rulezi <code>docker stats<\/code> ca s\u0103 vezi cine-i cel mai gras consumator activ.<\/li>\n\n\n\n<li>Verifici limitele de RAM pe container. Dac\u0103 n-ai, pui una imediat \u00een Compose.<\/li>\n\n\n\n<li>Verifici dac\u0103 ai swap activ pe VPS. Dac\u0103 nu, faci un fi\u0219ier de 1-2 GB.<\/li>\n\n\n\n<li>Intri \u00een config-ul aplica\u021biei \u0219i tai din connection pools, limitezi heap-ul sau tai din workeri.<\/li>\n\n\n\n<li>Dac\u0103 ai f\u0103cut tot tuning-ul corect \u0219i containerul \u00eenc\u0103 sufer\u0103, scoate portofelul \u0219i mut\u0103-te pe un plan cu mai mult RAM.<\/li>\n<\/ol>\n\n\n\n<h2 class=\"wp-block-heading\">Gre\u0219eli fatale (Ce s\u0103 NU faci)<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Nu opri sub nicio form\u0103 OOM killer-ul. El e acolo fix ca s\u0103 nu-\u021bi crape kernel-ul complet. F\u0103r\u0103 el, tot VPS-ul va face freeze.<\/li>\n\n\n\n<li>Nu pune <code>restart: always<\/code> la nimereal\u0103 pe containere f\u0103r\u0103 limite de memorie. Dac\u0103 containerul ia OOM kill \u0219i se restarteaz\u0103 de 50 de ori pe or\u0103, \u00ee\u021bi toac\u0103 CPU-ul \u0219i I\/O-ul discului scriind gigabytes de log-uri de eroare.<\/li>\n\n\n\n<li>Nu crea un fi\u0219ier de swap de 10 GB pe un server de 2 GB. \u00ce\u021bi vei transforma SSD-ul \u00eentr-un bottleneck oribil \u0219i serverul va deveni lent ca un melc f\u0103r\u0103 s\u0103 pice efectiv.<\/li>\n\n\n\n<li>Nu ignora exit code-ul 137 crez\u00e2nd c\u0103 e \u201eo fi\u021b\u0103 de la Docker\u201d. Este confirmarea strict\u0103 matematic\u0103 a faptului c\u0103 \u021bi-ai dep\u0103\u0219it resursele.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">Checklist pentru preven\u021bie pe termen lung<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Seteaz\u0103 limite de memorie pe 100% din containerele expuse \u00een produc\u021bie.<\/li>\n\n\n\n<li>Seteaz\u0103 limite de CPU dac\u0103 ai containere care compileaz\u0103 cod sau fac proces\u0103ri masive de tip batch.<\/li>\n\n\n\n<li>Tuneaz\u0103 parametrii din baze de date ca s\u0103 nu dep\u0103\u0219easc\u0103 memoria fizic\u0103 disponibil\u0103 a host-ului.<\/li>\n\n\n\n<li>Pentru orice aplica\u021bie Java, declar\u0103 explicit m\u0103rimea de heap \u00een environment variables.<\/li>\n\n\n\n<li>Arunc\u0103 mereu 1-2 GB de swap pentru plasa de siguran\u021b\u0103 la spike-uri.<\/li>\n\n\n\n<li>Stai cu ochii pe <code>docker stats<\/code> primele ore dup\u0103 un deployment nou.<\/li>\n\n\n\n<li>Las\u0103 obligatoriu un buffer de 20-30% din RAM liber strict pentru overhead-ul sistemului de operare \u0219i al Docker daemon-ului.<\/li>\n\n\n\n<li>Ai grij\u0103 cu ce unealt\u0103 faci backup. <a href=\"https:\/\/www.serverspan.ro\/blog\/borgbackup-vs-restic-care-iti-protejeaza-de-fapt-datele-de-pe-vps\/\">Compara\u021bia BorgBackup vs Restic<\/a> \u00ee\u021bi arat\u0103 clar care dintre ele \u0219tie s\u0103 gestioneze corect memoria pe servere mici f\u0103r\u0103 s\u0103 cear\u0103 OOM kill \u00een timpul proceselor masive de deduplicare.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">C\u00e2nd solu\u021bia este pur \u0219i simplu hardware mai puternic<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Ajungi rapid la un prag la care niciun tutorial de tuning nu te mai ajut\u0103. Dac\u0103 \u00eencerci s\u0103 rulezi pe acela\u0219i VPS de 2 GB un Nextcloud (care genereaz\u0103 cache video), un Immich (care ruleaz\u0103 scan\u0103ri ML de fa\u021b\u0103) \u0219i o instan\u021b\u0103 grea de PostgreSQL, faci echilibristic\u0103 pe ghea\u021b\u0103. Pur \u0219i simplu agregatele cer mai mult hardware.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Pentru stack-uri care con\u021bin elemente de OCR, Java masiv sau baze de date grele, instan\u021ba minim\u0103 recomandat\u0103 este un <a href=\"https:\/\/www.serverspan.ro\/virtual-servers\">VPS KVM ServerSpan<\/a> din clasa ct.Steady (4 Core, 4 GB RAM, 50 GB SSD). Trecerea la hardware superior este singura op\u021biune sustenabil\u0103 c\u00e2nd ai plafonat eficienta optimiz\u0103rilor software.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Costul unui upgrade mic de VPS se amortizeaz\u0103 instant dac\u0103 pui \u00een balan\u021b\u0103 orele pierdute citind loguri, coruperea datelor la fiecare hard kill luat de baza de date \u0219i downtime-ul aferent. Urmeaz\u0103 pa\u0219ii \u00een ordinea logic\u0103: 1. Diagnosticheaz\u0103 clar. 2. Tuneaz\u0103 configura\u021biile. 3. M\u0103re\u0219te resursele fizice (Scale up).<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Iar dac\u0103 vrei o perspectiv\u0103 mai tehnic\u0103 despre cum s\u0103 evi\u021bi picajele \u00een produc\u021bie cauzate strict de neglijen\u021ba \u00een administrare, cite\u0219te <a href=\"https:\/\/www.serverspan.ro\/blog\/actualizarile-automate-oarbe-pentru-docker-nu-inseamna-mentenanta-docker-29-a-facut-asta-imposibil-de-ignorat\/\">analiza incidentului Docker 29<\/a> \u2014 o radiografie clar\u0103 despre cum auto-update-urile la orb, healthcheck-urile ignorate \u0219i scripturile netestate duc fix la acelea\u0219i downtime-uri penibile.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Containerul t\u0103u a rulat brici s\u0103pt\u0103m\u00e2ni la r\u00e2nd. Apoi \u00eencepe brusc s\u0103 se restarteze la fiecare c\u00e2teva ore. Te ui\u021bi \u00een loguri \u0219i e bezn\u0103 total\u0103, nicio eroare util\u0103. Dai un docker ps \u0219i vezi c\u0103 statusul face ping-pong \u00eentre Up \u0219i Restarting. Singurul indiciu real este un exit code 137, care \u00eenseamn\u0103 c\u0103 procesul [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":2898,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[9,12],"tags":[16447,16435,16460,16450,16453,16462,16443,16455,16439,16454,16452,16440,16459,16461,16438,16456,16444,16446,16451,16445,16436,16441,16449,16437,13971,16442,16458,16448,16457],"class_list":["post-2900","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tutorials","category-virtual-servers","tag-container-exit-137-fix","tag-container-keeps-restarting","tag-container-killed-by-kernel","tag-container-memory-leak","tag-container-restart-loop-fix","tag-docker-build-oom-killed","tag-docker-cgroup-memory","tag-docker-compose-memory-limit","tag-docker-container-crashed","tag-docker-heap-size-java","tag-docker-memory-diagnostic","tag-docker-memory-limit-vps","tag-docker-memory-pressure-alert","tag-docker-oom-killed-postgres","tag-docker-oom-killed-vps","tag-docker-out-of-memory-fix","tag-docker-production-troubleshooting","tag-docker-ram-usage","tag-docker-stats-memory","tag-docker-swap-configuration","tag-exit-code-137-docker","tag-fix-docker-oom","tag-linux-oom-killer-explained","tag-oom-killer-linux-vps","tag-vps-docker-performance","tag-vps-memory-full","tag-vps-oom-killer-disable","tag-vps-resource-limits","tag-vps-swap-file-docker"],"_links":{"self":[{"href":"https:\/\/www.serverspan.com\/blog\/index.php?rest_route=\/wp\/v2\/posts\/2900","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.serverspan.com\/blog\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.serverspan.com\/blog\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.serverspan.com\/blog\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.serverspan.com\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=2900"}],"version-history":[{"count":2,"href":"https:\/\/www.serverspan.com\/blog\/index.php?rest_route=\/wp\/v2\/posts\/2900\/revisions"}],"predecessor-version":[{"id":2902,"href":"https:\/\/www.serverspan.com\/blog\/index.php?rest_route=\/wp\/v2\/posts\/2900\/revisions\/2902"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.serverspan.com\/blog\/index.php?rest_route=\/wp\/v2\/media\/2898"}],"wp:attachment":[{"href":"https:\/\/www.serverspan.com\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=2900"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.serverspan.com\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=2900"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.serverspan.com\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=2900"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}