Merge pull request #752 from jberkenbilt/report-mem-usage

Report mem usage

Merge pull request #752 from jberkenbilt/report-mem-usage
Report mem usage
Jay Berkenbilt · GitHub
2 parents 7b3134ef 4aac7c32
Showing 19 changed files with 176 additions and 10 deletions
TODO
cSpell.json
include/qpdf/QPDFJob.hh
include/qpdf/QUtil.hh
include/qpdf/auto_job_c_main.hh
job.sums
job.yml
libqpdf/CMakeLists.txt
libqpdf/QPDFJob.cc
libqpdf/QPDFJob_config.cc
libqpdf/QUtil.cc
libqpdf/qpdf/auto_job_help.hh
libqpdf/qpdf/auto_job_init.hh
libqpdf/qpdf/auto_job_json_init.hh
libqpdf/qpdf/auto_job_schema.hh
libqpdf/qpdf/qpdf-config.h.in
libtests/qtest/qutil/qutil.out
libtests/qutil.cc
manual/cli.rst
@@ -4,10 +4,12 @@ Next
  
 Before Release:
  
-* Review in order #729, #726, #747
+* Review in order #726
 * Make ./performance_check usable by other people by having published
   files to use for testing.
   * https://opensource.adobe.com/dc-acrobat-sdk-docs/standards/pdfstandards/pdf/PDF32000_2008.pdf
+* Incorporate --report-mem-usage into performance testing. Make sure
+  there is some test somewhere that exercises the millions of nulls case.
 * Evaluate issues tagged with `next`
 * Stay on top of https://github.com/pikepdf/pikepdf/pull/315
  
@@ -131,6 +131,7 @@
     "esize",
     "eval",
     "extlibdir",
+    "fclose",
     "fdict",
     "ffield",
     "fghij",
@@ -268,6 +269,7 @@
     "maxdepth",
     "maxobjectid",
     "mdash",
+    "memstream",
     "mindepth",
     "mkdir",
     "mkinstalldirs",
@@ -711,6 +711,7 @@ class QPDFJob
         bool json_input;
         bool json_output;
         std::string update_from_json;
+        bool report_mem_usage;
     };
     std::shared_ptr<Members> m;
 };
@@ -525,7 +525,17 @@ namespace QUtil
         wchar_t const* const argv[],
         std::function<int(int, char const* const[])> realmain);
 #endif // QPDF_NO_WCHAR_T
-};     // namespace QUtil
+
+    // Try to return the maximum amount of memory allocated by the
+    // current process and its threads. Return 0 if unable to
+    // determine. This is Linux-specific and not implemented to be
+    // completely reliable. It is used during development for
+    // performance testing to detect changes that may significantly
+    // change memory usage. It is not recommended for use for other
+    // purposes.
+    QPDF_DLL
+    size_t get_max_memory_usage();
+}; // namespace QUtil
  
 inline bool
 QUtil::is_hex_digit(char ch)
@@ -33,6 +33,7 @@ QPDF_DLL Config* qdf();
 QPDF_DLL Config* rawStreamData();
 QPDF_DLL Config* recompressFlate();
 QPDF_DLL Config* removePageLabels();
+QPDF_DLL Config* reportMemUsage();
 QPDF_DLL Config* requiresPassword();
 QPDF_DLL Config* showEncryption();
 QPDF_DLL Config* showEncryptionKey();
@@ -3,15 +3,15 @@ generate_auto_job 9abe2ec994fb98526f5e3c0c199ce2e61a868463cb522a5bc6e9730b655341
 include/qpdf/auto_job_c_att.hh 4c2b171ea00531db54720bf49a43f8b34481586ae7fb6cbf225099ee42bc5bb4
 include/qpdf/auto_job_c_copy_att.hh 50609012bff14fd82f0649185940d617d05d530cdc522185c7f3920a561ccb42
 include/qpdf/auto_job_c_enc.hh 28446f3c32153a52afa239ea40503e6cc8ac2c026813526a349e0cd4ae17ddd5
-include/qpdf/auto_job_c_main.hh cdba1ae6ea5525a585d10a3dd95b7996d62b17de4211fe658b78d9d463b0f313
+include/qpdf/auto_job_c_main.hh 493b9798f5ff8bbcb07c0238693554d77eefa4ae71ce1a0d466de94e3a7a3966
 include/qpdf/auto_job_c_pages.hh b3cc0f21029f6d89efa043dcdbfa183cb59325b6506001c18911614fe8e568ec
 include/qpdf/auto_job_c_uo.hh ae21b69a1efa9333050f4833d465f6daff87e5b38e5106e49bbef5d4132e4ed1
-job.yml f9564f18b08a45d17328af43652645771d3498471820c858b8c9013a193e1412
+job.yml a6f22d425980ed960c77c0a4197f46924c14e7943358cd9f0b75811bb1c480ad
 libqpdf/qpdf/auto_job_decl.hh 7844eba58edffb9494b19e8eca6fd59a24d6e152ca606c3b07da569f753df2da
-libqpdf/qpdf/auto_job_help.hh 53306e4aef8aaca641c0087bc9e064ada1c44a94b826c0bcac7b4eb0c8c41fd5
-libqpdf/qpdf/auto_job_init.hh fd1635a5ad6ba16b7ae008467145560a59a5ecfd10d29c5ef7cd0d8347747cd2
+libqpdf/qpdf/auto_job_help.hh 3e9385a7e0dae993467647466fa30f30baa5968f9270c73ff4e664f5aa415dbe
+libqpdf/qpdf/auto_job_init.hh ccb881733849dff5c05721f1aa5c35447cedd415e881c3fef6573901e45be056
 libqpdf/qpdf/auto_job_json_decl.hh 06caa46eaf71db8a50c046f91866baa8087745a9474319fb7c86d92634cc8297
-libqpdf/qpdf/auto_job_json_init.hh 59545578a2e47c660ff98516ed53f06638be75eb4658e2a09d32cc08e0cb7268
-libqpdf/qpdf/auto_job_schema.hh 5352ef1be1ad7cc6f4f36dab88f2937d278e6bd3a0e2d46259794dc226c8ba6b
+libqpdf/qpdf/auto_job_json_init.hh 7ac8f42fb39eda56144ab62b30152a56e9bb2224d0596eb826b7bc421a78d26b
+libqpdf/qpdf/auto_job_schema.hh 17352791b09c3b8a8db766375cce31d70c98b67b44ecc398e2ac78984e34fe90
 manual/_ext/qpdf.py 6add6321666031d55ed4aedf7c00e5662bba856dfcd66ccb526563bffefbb580
-manual/cli.rst 41ee93f23f46160fe9eaf7c99fd2ab3bd2e0f6792a341a35bdac1a41cb853ed5
+manual/cli.rst e3fa48bb30c981df1f74d474887155cd6a46f9010b91cd1b7b57e582bf3bf877
@@ -127,6 +127,7 @@ options:
       - recompress-flate
       - remove-page-labels
       - replace-input
+      - report-mem-usage
       - requires-password
       - show-encryption
       - show-encryption-key
@@ -413,6 +414,7 @@ json:
       Pages.password:
       _range: "page range"
   remove-page-labels:
+  report-mem-usage:
   rotate:
   overlay:
     _file: "source file for overlay"
@@ -376,6 +376,29 @@ int main(int argc, char* argv[]) {
   endif()
 endfunction()
  
+check_c_source_compiles(
+"#include <malloc.h>
+#include <stdio.h>
+int main(int argc, char* argv[]) {
+    malloc_info(0, stdout);
+    return 0;
+}"
+    HAVE_MALLOC_INFO)
+
+check_c_source_compiles(
+"#include <stdio.h>
+#include <stdlib.h>
+int main(int argc, char* argv[]) {
+    char* buf;
+    size_t size;
+    FILE* f;
+    f = open_memstream(&buf, &size);
+    fclose(f);
+    free(buf);
+    return 0;
+}"
+    HAVE_OPEN_MEMSTREAM)
+
 qpdf_check_ll_fmt("%lld" fmt_lld)
 qpdf_check_ll_fmt("%I64d" fmt_i64d)
 qpdf_check_ll_fmt("%I64lld" fmt_i64lld)
@@ -417,7 +417,8 @@ QPDFJob::Members::Members() :
     check_is_encrypted(false),
     check_requires_password(false),
     json_input(false),
-    json_output(false)
+    json_output(false),
+    report_mem_usage(false)
 {
 }
  
@@ -625,6 +626,14 @@ QPDFJob::run()
                                      << ": operation succeeded with warnings\n";
         }
     }
+    if (m->report_mem_usage) {
+        // Call get_max_memory_usage before generating output. When
+        // debugging, it's easier if print statements from
+        // get_max_memory_usage are not interleaved with the output.
+        auto mem_usage = QUtil::get_max_memory_usage();
+        *this->m->log->getWarn()
+            << "qpdf-max-memory-usage " << mem_usage << "\n";
+    }
 }
  
 bool
@@ -503,6 +503,13 @@ QPDFJob::Config::removePageLabels()
 }
  
 QPDFJob::Config*
+QPDFJob::Config::reportMemUsage()
+{
+    o.m->report_mem_usage = true;
+    return this;
+}
+
+QPDFJob::Config*
 QPDFJob::Config::requiresPassword()
 {
     o.m->check_requires_password = true;
@@ -37,6 +37,9 @@
 # include <sys/stat.h>
 # include <unistd.h>
 #endif
+#ifdef HAVE_MALLOC_INFO
+# include <malloc.h>
+#endif
  
 // First element is 24
 static unsigned short pdf_doc_low_to_unicode[] = {
@@ -1968,3 +1971,73 @@ QUtil::call_main_from_wmain(
 }
  
 #endif // QPDF_NO_WCHAR_T
+
+size_t
+QUtil::get_max_memory_usage()
+{
+#if defined(HAVE_MALLOC_INFO) && defined(HAVE_OPEN_MEMSTREAM)
+    static std::regex tag_re("<(/?\\w+)([^>]*?)>");
+    static std::regex attr_re("(\\w+)=\"(.*?)\"");
+
+    char* buf;
+    size_t size;
+    FILE* f = open_memstream(&buf, &size);
+    if (f == nullptr) {
+        return 0;
+    }
+    malloc_info(0, f);
+    fclose(f);
+    if (QUtil::get_env("QPDF_DEBUG_MEM_USAGE")) {
+        fprintf(stderr, "%s", buf);
+    }
+
+    // Warning: this code uses regular expression to extract data from
+    // an XML string. This is generally a bad idea, but we're going to
+    // do it anyway because QUtil.hh warns against using this function
+    // for other than development/testing, and if this function fails
+    // to generate reasonable output during performance testing, it
+    // will be noticed.
+
+    // This is my best guess at how to interpret malloc_info. Anyway
+    // it seems to provide useful information for detecting code
+    // changes that drastically change memory usage.
+    size_t result = 0;
+    try {
+        std::cregex_iterator m_begin(buf, buf + size, tag_re);
+        std::cregex_iterator cr_end;
+        std::sregex_iterator sr_end;
+
+        int in_heap = 0;
+        for (auto m = m_begin; m != cr_end; ++m) {
+            std::string tag(m->str(1));
+            if (tag == "heap") {
+                ++in_heap;
+            } else if (tag == "/heap") {
+                --in_heap;
+            } else if (in_heap == 0) {
+                std::string rest = m->str(2);
+                std::map<std::string, std::string> attrs;
+                std::sregex_iterator a_begin(rest.begin(), rest.end(), attr_re);
+                for (auto m2 = a_begin; m2 != sr_end; ++m2) {
+                    attrs[m2->str(1)] = m2->str(2);
+                }
+                if (tag == "total") {
+                    if (attrs.count("size") > 0) {
+                        result += QIntC::to_size(
+                            QUtil::string_to_ull(attrs["size"].c_str()));
+                    }
+                } else if (tag == "system" && attrs["type"] == "max") {
+                    result += QIntC::to_size(
+                        QUtil::string_to_ull(attrs["size"].c_str()));
+                }
+            }
+        }
+    } catch (...) {
+        // ignore -- just return 0
+    }
+    free(buf);
+    return result;
+#else
+    return 0;
+#endif
+}
@@ -883,6 +883,9 @@ for debugging qpdf.
 ap.addOptionHelp("--test-json-schema", "testing", "test generated json against schema", R"(This is used by qpdf's test suite to check consistency between
 the output of qpdf --json and the output of qpdf --json-help.
 )");
+ap.addOptionHelp("--report-mem-usage", "testing", "best effort report of memory usage", R"(This is used by qpdf's performance test suite to report the
+maximum amount of memory used in supported environments.
+)");
 }
 static void add_help(QPDFArgParser& ap)
 {
@@ -69,6 +69,7 @@ this-&gt;ap.addBare(&quot;raw-stream-data&quot;, [this](){c_main-&gt;rawStreamData();});
 this->ap.addBare("recompress-flate", [this](){c_main->recompressFlate();});
 this->ap.addBare("remove-page-labels", [this](){c_main->removePageLabels();});
 this->ap.addBare("replace-input", b(&ArgParser::argReplaceInput));
+this->ap.addBare("report-mem-usage", [this](){c_main->reportMemUsage();});
 this->ap.addBare("requires-password", [this](){c_main->requiresPassword();});
 this->ap.addBare("show-encryption", [this](){c_main->showEncryption();});
 this->ap.addBare("show-encryption-key", [this](){c_main->showEncryptionKey();});
@@ -409,6 +409,9 @@ popHandler(); // key: pages
 pushKey("removePageLabels");
 addBare([this]() { c_main->removePageLabels(); });
 popHandler(); // key: removePageLabels
+pushKey("reportMemUsage");
+addBare([this]() { c_main->reportMemUsage(); });
+popHandler(); // key: reportMemUsage
 pushKey("rotate");
 addParameter([this](std::string const& p) { c_main->rotate(p); });
 popHandler(); // key: rotate
@@ -144,6 +144,7 @@ static constexpr char const* JOB_SCHEMA_DATA = R&quot;({
     }
   ],
   "removePageLabels": "remove explicit page numbers",
+  "reportMemUsage": "best effort report of memory usage",
   "rotate": "rotate pages",
   "overlay": {
     "file": "source file for overlay",
@@ -21,6 +21,8 @@
 #cmakedefine HAVE_LOCALTIME_R 1
 #cmakedefine HAVE_RANDOM 1
 #cmakedefine HAVE_TM_GMTOFF 1
+#cmakedefine HAVE_MALLOC_INFO 1
+#cmakedefine HAVE_OPEN_MEMSTREAM 1
  
 /* printf format for long long */
 #cmakedefine LL_FMT "${LL_FMT}"
@@ -134,3 +134,5 @@ D:20210209191925Z
 2021-02-09T19:19:25Z
 ---- is_long_long
 done
+---- memory usage
+memory usage okay
@@ -703,6 +703,18 @@ is_long_long_test()
     std::cout << "done" << std::endl;
 }
  
+void
+memory_usage_test()
+{
+    auto u1 = QUtil::get_max_memory_usage();
+    if (u1 > 0) {
+        auto x = QUtil::make_shared_array<int>(10 << 20);
+        auto u2 = QUtil::get_max_memory_usage();
+        assert(u2 > u1);
+    }
+    std::cout << "memory usage okay" << std::endl;
+}
+
 int
 main(int argc, char* argv[])
 {
@@ -739,6 +751,8 @@ main(int argc, char* argv[])
         timestamp_test();
         std::cout << "---- is_long_long" << std::endl;
         is_long_long_test();
+        std::cout << "---- memory usage" << std::endl;
+        memory_usage_test();
     } catch (std::exception& e) {
         std::cout << "unexpected exception: " << e.what() << std::endl;
     }
@@ -3463,6 +3463,16 @@ Related Options
    memory and is therefore unsuitable for use with large files. This
    is why it's also not on by default.
  
+.. qpdf:option:: --report-mem-usage
+
+   .. help: best effort report of memory usage
+
+      This is used by qpdf's performance test suite to report the
+      maximum amount of memory used in supported environments.
+
+   This is used by qpdf's performance test suite to report the maximum
+   amount of memory used in supported environments.
+
 .. _unicode-passwords:
  
 Unicode Passwords