Fix crashes in gsymutil resource accounting

Summary:
poll uses wnohang option, so ocasionally we got empty stats and the service crashed because we never actually waited for the process to complete and there is a safeguard in folly for that https://fburl.com/code/ujrzktm5 We mostly succeeded probably because communicate call above returned shortly before the process death, but there were ocasional races.

We could poll in a cycle in the service until the state is running, but that seemed like a cludge. So I decided to make changes to folly.

I thought about adding options field to poll. But semantically seemed wrong that we can wait in a function named poll. Also there is no retry loop for properly handling waiting erros.

Wait already uses waitpid under the hood, changing it to wait4 seemed risky.

So I decided to add a new function which is kinda of a mix of a two.

Reviewed By: jwiepert, hodgesds

Differential Revision: D73506301

fbshipit-source-id: 344833608409a5bdd5873e0fe50b863b4347c1a5

Author: peremyach

third-party/folly/src/folly/Subprocess.cpp

@@ -1085,6 +1085,25 @@ ProcessReturnCode Subprocess::wait() {
   return returnCode_;
 }
 
+ProcessReturnCode Subprocess::waitAndGetRusage(struct rusage* ru) {
+  returnCode_.enforce(ProcessReturnCode::RUNNING);
+  DCHECK_GT(pid_, 0);
+  int status;
+  pid_t found;
+  do {
+    found = ::wait4(pid_, &status, 0, ru);
+  } while (found == -1 && errno == EINTR);
+  // The only two remaining errors are ECHILD (other code reaped the
+  // child?), or EINVAL (cosmic rays?), and both merit an abort:
+  PCHECK(found != -1) << "wait4(" << pid_ << ", &status, 0, resourceUsage)";
+  // Though the child process had quit, this call does not close the pipes
+  // since its descendants may still be using them.
+  DCHECK_EQ(found, pid_);
+  returnCode_ = ProcessReturnCode::make(status);
+  pid_ = -1;
+  return returnCode_;
+}
+
 void Subprocess::waitChecked() {
   wait();
   checkStatus(returnCode_);

third-party/folly/src/folly/Subprocess.h

@@ -670,6 +670,14 @@ class Subprocess {
    */
   ProcessReturnCode wait();
 
+  /**
+   * Wait for the process to terminate and return its status and rusage.  Like
+   * poll(), the only exception this can throw is std::logic_error if you call
+   * this on a Subprocess whose status is not RUNNING.  Aborts on egregious
+   * violations of contract, like an out-of-band wait4(p.pid(), 0, 0, nullptr).
+   */
+  ProcessReturnCode waitAndGetRusage(struct rusage* ru);
+
   /**
    * Wait for the process to terminate, throw if unsuccessful.
    */