nodejs / nodejs/node

Running a process with DevTools debugger attached leaks memory

Ouverte
#39,063 4 commentaires 13 réactions 0 personnes assignées Voir sur GitHub

Personne n'a encore pris cette issue.

inspector memory
Langage dominant
JavaScript
Étoiles
122k
Forks
37.3k
Merge moyen
4 j 2 h
PR mergées (30 j)
283

Description

While the Chromium DevTools debugger is attached, RSS memory usage of a Node process keeps increasing without being collected. Once the debugger is detached, the memory usage flattens, however the memory is never released. When the debugger is not attached to the process in its runtime, memory allocation occurs as expected. It is entirely possible that this might not be related to Node (v8, perhaps) - however, this issue might be a good place to start a discussion.

Inspecting htop on a Raspberry Pi 3B machine:

  • Process after running for ~24 hours without a debugger attached: stable at around ~180MB RSS:
    no debugger

  • The same process, after attaching the debugger for ~15 minutes (with no actual user debugging active nor any logs being displayed): RSS at around ~245MB and rising:
    debugger

  • After disconnecting the debugger (removing the connection, closing and re-opening the debugger window), RSS stabilizes at around ~260MB. After ~30 minutes of further activity, the RSS usage seems stable; however, the memory is not freed:
    after

After running valgrind for various amounts of time on a different Linux machine, there seems to be an excessive amount of v8_inspector::AsyncStackTrace::capture calls. Using the Disable async stack traces option in DevTools > Settings > Debugger seems to reduce the leak somewhat, however, this wasn't reproducible across machines. Furthermore, testing the same process on Windows 10 in the same environment leads to no memory leaks.

Attached are valgrind dumps for various tested nodejs versions. Note that each dump was created with the process running for differing amounts, hence the total memory leaked will not be as relevant.

valgrind.zip

What steps will reproduce the bug?
  1. Set up a Node process, possibly with asynchronous stack activity
  2. Attach a Chrome DevTools debugger instance
  3. Observe a memory leak as long as the debugger is attached
How often does it reproduce? Is there a required condition?
  • Only seems to reproduce on Linux, not Windows
  • Might be related to async stack traces
  • At first, seemed to be related to the mariadb NPM module - but upon further testing with the module inactive, the memory leak kept occuring
What is the expected behavior?

No memory leak should occur as the debugger is being used.

What do you see instead?

Depending on the process being run, upwards of hundreds of MB of RAM will leak in a couple of minutes as the debugger is attached.

Additional information

A similar type of memory leak seems to be present in versions prior to 14.0, however, in a much smaller scale, and occuring much more slowly. Reference: nodejs/node#28787, possibly nodejs/node#28420

Guide de contribution

Ouvrir le guide de contribution

Par où commencer

  1. Lisez l'issue en entier, puis le guide de contribution du projet.
  2. Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
  3. Forkez le dépôt et travaillez sur une branche.
  4. Ouvrez une pull request qui référence le numéro de l'issue.

Piste de recherche

Commencez par reproduire la fuite en suivant les étapes indiquées sous Linux, avec Chrome DevTools attaché, et comparez les exécutions avec les async stack traces activées et désactivées. Examinez les dumps Valgrind fournis ainsi que l’activité excessive de v8_inspector::AsyncStackTrace::capture ; le travail est considéré comme terminé lorsque l’utilisation du débogueur ne provoque plus de croissance persistante de RSS ni de mémoire non libérée.

Rédigé par le modèle d'indexation à partir du texte de l'issue.

Évaluation

Stack technique
javascript, nodejs
Domaine
performance, tooling
Type d'issue
Bug
Difficulté
4/5
Temps estimé
3-5 jours
Activité
Calme
Clarté
Plutôt claire
Accessibilité débutants
45/100

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.