Option to revert to sequential read (Chia-Network#296)

* Option to revert to sequential read

* Reverse disable boolean flow

* Fix test

* debug ci runner

* debugging runner

* Add swap space to runner

* remove prints

* increase swap size

.github/workflows/build-test-cplusplus.yml

@@ -12,6 +12,7 @@ jobs:
         os: [ubuntu-20.04, windows-latest]
 
     steps:
+
     - name: Cancel previous runs on the same branch
       if: ${{ github.ref != 'refs/heads/main' }}
       uses: styfle/[email protected]
@@ -35,11 +36,16 @@ jobs:
       run: |
         sudo apt update
         sudo apt-get install valgrind -y
+        sudo fallocate -l 16G /swapfile
+        sudo chmod 600 /swapfile
+        sudo mkswap /swapfile
+        sudo swapon /swapfile
         mkdir build
         cd build
         cmake ../
         cmake --build . -- -j 6
         ctest -j 6 --output-on-failure
+        swapon -s
         valgrind --leak-check=full --show-leak-kinds=all --errors-for-leak-kinds=all ctest -j 6 --output-on-failure
 
     - name: cmake, RunTests with address- and undefined sanitizer on Ubuntu
@@ -49,6 +55,7 @@ jobs:
         cd build-asan
         cmake -DCMAKE_BUILD_TYPE=ASAN ../
         cmake --build . -- -j 6
+        swapon -s
         ./RunTests
 
     - name: cmake, RunTests with thread sanitizer on Ubuntu
@@ -58,4 +65,5 @@ jobs:
         cd build-tsan
         cmake -DCMAKE_BUILD_TYPE=TSAN ../
         cmake --build . -- -j 6
+        swapon -s
         ./RunTests

python-bindings/chiapos.cpp

@@ -119,19 +119,19 @@ PYBIND11_MODULE(chiapos, m)
                 delete[] quality_buf;
                 return ret;
             })
-        .def("get_full_proof", [](DiskProver &dp, const py::bytes &challenge, uint32_t index) {
+        .def("get_full_proof", [](DiskProver &dp, const py::bytes &challenge, uint32_t index, bool parallel_read) {
             std::string challenge_str(challenge);
             const uint8_t *challenge_ptr = reinterpret_cast<const uint8_t *>(challenge_str.data());
             py::gil_scoped_release release;
-            LargeBits proof = dp.GetFullProof(challenge_ptr, index);
+            LargeBits proof = dp.GetFullProof(challenge_ptr, index, parallel_read);
             py::gil_scoped_acquire acquire;
             uint8_t *proof_buf = new uint8_t[Util::ByteAlign(64 * dp.GetSize()) / 8];
             proof.ToBytes(proof_buf);
             py::bytes ret = py::bytes(
                 reinterpret_cast<char *>(proof_buf), Util::ByteAlign(64 * dp.GetSize()) / 8);
             delete[] proof_buf;
             return ret;
-        });
+        },py::arg("challenge"), py::arg("index"), py::arg("parallel_read") = true);
 
     py::class_<Verifier>(m, "Verifier")
         .def(py::init<>())

src/cli.cpp

@@ -83,6 +83,7 @@ int main(int argc, char *argv[]) try {
     string id = "022fb42c08c12de3a6af053880199806532e79515f94e83461612101f9412f9e";
     bool nobitfield = false;
     bool show_progress = false;
+    bool parallel_read = true;
     uint32_t buffmegabytes = 0;
 
     options.allow_unrecognised_options().add_options()(
@@ -102,6 +103,8 @@ int main(int argc, char *argv[]) try {
         cxxopts::value<uint32_t>(buffmegabytes))(
         "p, progress", "Display progress percentage during plotting",
         cxxopts::value<bool>(show_progress))(
+        "parallel_read", "Set to false to use sequential reads",
+        cxxopts::value<bool>(parallel_read)->default_value("true"))(
         "help", "Print help");
 
     auto result = options.parse(argc, argv);
@@ -177,7 +180,7 @@ int main(int argc, char *argv[]) try {
             for (uint32_t i = 0; i < qualities.size(); i++) {
                 k = prover.GetSize();
                 uint8_t *proof_data = new uint8_t[8 * k];
-                LargeBits proof = prover.GetFullProof(challenge_bytes, i);
+                LargeBits proof = prover.GetFullProof(challenge_bytes, i, parallel_read);
                 proof.ToBytes(proof_data);
                 cout << "Proof: 0x" << Util::HexStr(proof_data, k * 8) << endl;
                 delete[] proof_data;
@@ -259,7 +262,7 @@ int main(int argc, char *argv[]) try {
                 vector<LargeBits> qualities = prover.GetQualitiesForChallenge(hash.data());
 
                 for (uint32_t i = 0; i < qualities.size(); i++) {
-                    LargeBits proof = prover.GetFullProof(hash.data(), i);
+                    LargeBits proof = prover.GetFullProof(hash.data(), i, parallel_read);
                     uint8_t *proof_data = new uint8_t[proof.GetSize() / 8];
                     proof.ToBytes(proof_data);
                     cout << "i: " << num << std::endl;

src/prover_disk.hpp

@@ -201,7 +201,7 @@ class DiskProver {
     // Given a challenge, and an index, returns a proof of space. This assumes GetQualities was
     // called, and there are actually proofs present. The index represents which proof to fetch,
     // if there are multiple.
-    LargeBits GetFullProof(const uint8_t* challenge, uint32_t index)
+    LargeBits GetFullProof(const uint8_t* challenge, uint32_t index, bool parallel_read = true)
     {
         LargeBits full_proof;
 
@@ -219,7 +219,12 @@ class DiskProver {
             }
 
             // Gets the 64 leaf x values, concatenated together into a k*64 bit string.
-            std::vector<Bits> xs = GetInputs(p7_entries[index], 6);
+            std::vector<Bits> xs;
+            if (parallel_read) {
+                xs = GetInputs(p7_entries[index], 6);
+            } else {
+                xs = GetInputs(p7_entries[index], 6, &disk_file); // Passing in a disk_file disabled the parallel reads
+            }
 
             // Sorts them according to proof ordering, where
             // f1(x0) m= f1(x1), f2(x0, x1) m= f2(x2, x3), etc. On disk, they are not stored in
@@ -635,11 +640,18 @@ class DiskProver {
     // all of the leaves (x values). For example, for depth=5, it fetches the position-th
     // entry in table 5, reading the two back pointers from the line point, and then
     // recursively calling GetInputs for table 4.
-    std::vector<Bits> GetInputs(uint64_t position, uint8_t depth)
+    std::vector<Bits> GetInputs(uint64_t position, uint8_t depth, std::ifstream* disk_file = nullptr)
     {
-        // Create individual file handles to allow parallel processing
-        std::ifstream disk_file(filename, std::ios::in | std::ios::binary);
-        uint128_t line_point = ReadLinePoint(disk_file, depth, position);
+        uint128_t line_point;
+
+        if (!disk_file) {
+            // No disk file passed in, so we assume here we are doing parallel reads
+            // Create individual file handles to allow parallel processing
+            std::ifstream disk_file_parallel(filename, std::ios::in | std::ios::binary);
+            line_point = ReadLinePoint(disk_file_parallel, depth, position);
+        } else {
+            line_point = ReadLinePoint(*disk_file, depth, position);
+        }
         std::pair<uint64_t, uint64_t> xy = Encoding::LinePointToSquare(line_point);
 
         if (depth == 1) {
@@ -649,14 +661,22 @@ class DiskProver {
             ret.emplace_back(xy.first, k);   // x
             return ret;
         } else {
-            auto left_fut=std::async(std::launch::async, &DiskProver::GetInputs,this, (uint64_t)xy.second, (uint8_t)(depth - 1));
-            auto right_fut=std::async(std::launch::async, &DiskProver::GetInputs,this, (uint64_t)xy.first, (uint8_t)(depth - 1));
-            std::vector<Bits> left = left_fut.get();  // y
-            std::vector<Bits> right = right_fut.get();  // x
+            std::vector<Bits> left, right;
+            if (!disk_file) {
+                // no disk_file, so we do parallel reads here
+                auto left_fut=std::async(std::launch::async, &DiskProver::GetInputs,this, (uint64_t)xy.second, (uint8_t)(depth - 1), nullptr);
+                auto right_fut=std::async(std::launch::async, &DiskProver::GetInputs,this, (uint64_t)xy.first, (uint8_t)(depth - 1), nullptr);
+                left = left_fut.get();  // y
+                right = right_fut.get();  // x
+            } else {
+                left = GetInputs(xy.second, depth - 1, disk_file);  // y
+                right = GetInputs(xy.first, depth - 1, disk_file);  // x  
+            }
             left.insert(left.end(), right.begin(), right.end());
             return left;
         }
     }
+
 };
 
 #endif  // SRC_CPP_PROVER_DISK_HPP_

tests/test.cpp

@@ -509,6 +509,7 @@ void TestProofOfSpace(
         picosha2::hash256(hash_input.begin(), hash_input.end(), hash.begin(), hash.end());
         vector<LargeBits> qualities = prover.GetQualitiesForChallenge(hash.data());
         Verifier verifier = Verifier();
+
         for (uint32_t index = 0; index < qualities.size(); index++) {
             LargeBits proof = prover.GetFullProof(hash.data(), index);
             proof.ToBytes(proof_data);

tests/test_python_bindings.py

@@ -64,6 +64,7 @@ def test_k_21(self):
         pr = DiskProver(str(Path("myplot.dat")))
 
         total_proofs: int = 0
+        total_proofs2: int = 0
         iterations: int = 5000
 
         v = Verifier()
@@ -79,11 +80,25 @@ def test_k_21(self):
                 )
                 assert computed_quality == quality
                 total_proofs += 1
+            for index, quality in enumerate(pr.get_qualities_for_challenge(challenge)):
+                proof = pr.get_full_proof(challenge, index, parallel_read=False)
+                assert len(proof) == 8 * pr.get_size()
+                computed_quality = v.validate_proof(
+                    plot_seed, pr.get_size(), challenge, proof
+                )
+                assert computed_quality == quality
+                total_proofs2 += 1
 
         print(
             f"total proofs {total_proofs} out of {iterations}\
             {total_proofs / iterations}"
         )
+        print(
+            f"total proofs (sequential reads) {total_proofs2} out of {iterations}\
+            {total_proofs2 / iterations}"
+        )
+
+        assert total_proofs2 == total_proofs
         assert total_proofs > 4000
         assert total_proofs < 6000
         pr = None
@@ -125,7 +140,9 @@ def test_faulty_plot_doesnt_crash(self):
         all_data = bytearray(f.read())
         f.close()
         assert len(all_data) > 20000000
-        all_data_bad = all_data[:20000000] + bytearray(token_bytes(10000)) + all_data[20100000:]
+        all_data_bad = (
+            all_data[:20000000] + bytearray(token_bytes(10000)) + all_data[20100000:]
+        )
         f_bad = open("myplotbad.dat", "wb")
         f_bad.write(all_data_bad)
         f_bad.close()
@@ -141,7 +158,9 @@ def test_faulty_plot_doesnt_crash(self):
                 print(i)
             challenge = sha256(i.to_bytes(4, "big")).digest()
             try:
-                for index, quality in enumerate(pr.get_qualities_for_challenge(challenge)):
+                for index, quality in enumerate(
+                    pr.get_qualities_for_challenge(challenge)
+                ):
                     proof = pr.get_full_proof(challenge, index)
                     computed_quality = v.validate_proof(
                         plot_id, pr.get_size(), challenge, proof