File descriptor leak

[AlexHls]

Describe the bug

Sampling a model leaks file descriptors, which becomes a problem when running many fits in sequence using the same process (e.g. during bootstrap resampling). Increasing the number of possible open files is a possible temporary fix, but does not work for huge number of resampling sequences. In my case, I need to run ~510^6 fits, which inevitably leads to and OSError [Errno 24]. It seems to originate from some bad file descriptors left behind by the `/tmp/httpstan_.sockfiles and some sockets (i.e. when trying to manually close the file descriptors,os.fstat` returns a BadFileDescriptor error). I've attached a MWE illustarting the increase in the number of used file descriptors. This might be an upstream problem, but I could not verify this.

Describe your system

• Operating system: EndeavourOS (Arch-based)
• CPU type: x86-64
• C++ compiler version: GCC 12.2

  import stan
  import os
  
  model = """
  data {
    int<lower = 0> N;
    vector[N] x;
    vector[N] y;
    int<lower = 0, upper = 1> resample;
  }
  transformed data {
    simplex[N] uniform = rep_vector(1.0 / N, N);
    int<lower = 1, upper = N> boot_idxs[N];
    for (n in 1:N)
      boot_idxs[n] = resample ? categorical_rng(uniform) : n;
  }
  parameters {
    real alpha;
    real beta;
    real<lower = 0> sigma;
  }
  model {
    y[boot_idxs] ~ normal(alpha + beta * x[boot_idxs], sigma);
  }
  """
  
  schools_data = {"J": 8,
                  "y": [28,  8, -3,  7, -1,  1, 18, 12],
                  "sigma": [15, 10, 16, 11,  9, 11, 10, 18]}
  data = {
      "N": 4,
      "x": [28, 8, -3, 7],
      "y":[15, 10, 16, 11],
      "resample": 1,
  }
  
  def get_fds():
      return len(set(os.listdir('/proc/self/fd')))
  
  print("File descriptors before build: %d" % get_fds())
  posterior = stan.build(model, data=data)
  print("File descriptors before fit: %d" % get_fds())
  fit = posterior.sample(num_chains=2, num_samples=1000)
  print("File descriptors before second fit: %d" % get_fds())
  fit = posterior.sample(num_chains=2, num_samples=1000)
  print("File descriptors after second fit: %d" % get_fds())
  

  Output (without stanc messages):

  File descriptors before build: 10
  Building: found in cache, done.
  File descriptors before fit: 10
  Sampling: 100% (4000/4000), done.
  File descriptors before second fit: 28
  Sampling: 100% (4000/4000), done.
  File descriptors after second fit: 30

[riddell-stan]

Thanks for the investigation.

What version of httpstan are you using? This should have been fixed in https://github.com/stan-dev/httpstan/pull/614

[AlexHls]

Apparently my httpstan version did not update when updating pystan. Updating manually to the newest version of httpstan seems to fix the issue. Thanks for the hint.