mirror of
https://github.com/dashpay/dash.git
synced 2024-12-25 12:02:48 +01:00
Merge #12495: Increase LevelDB max_open_files
ccedbaf
Increase LevelDB max_open_files unless on 32-bit Unix. (Evan Klitzke)
Pull request description:
Currently we set `max_open_files = 64` on all architectures due to concerns about file descriptor exhaustion. This is extremely expensive due to how LevelDB is designed.
When a LevelDB file handle is opened, a bloom filter and block index are decoded, and some CRCs are checked. Bloom filters and block indexes in open table handles can be checked purely in memory. This means that when doing a key lookup, if a given table file may contain a given key, all of the lookup operations can happen completely in RAM until the block itself is fetched. In the common case fetching the block is one disk seek, because the block index stores its physical offset. This is the ideal case, and what we want to happen as often as possible.
If a table file handle is not open in the table cache, then in addition to the regular system calls to open the file, the block index and bloom filter need to be decoded before they can be checked. This is expensive and is something we want to avoid.
The current setting of 64 file handles means that on a synced node, only about 4% of key lookups can be satisifed by table file handles that are actually open and in memory.
The original concerns about file descriptor exhaustion are unwarranted on most systems because:
* On 64-bit POSIX hosts LevelDB will open up to 1000 file descriptors using `mmap()`, and it does not retain an open file descriptor for such files.
* On Windows non-socket files do not interfere with the main network `select()` loop, so the same fd exhaustion issues do not apply there.
This change keeps the default `max_open_files` value (which is 1000) on all systems except 32-bit POSIX hosts (which do not use `mmap()`). Open file handles use about 20 KB of memory (for the block index), so the extra file handles do not cause much memory overhead. At most 1000 will be open, and a fully synced node right now has about 1500 such files.
Profile of `loadblk` thread before changes: https://monad.io/maxopenfiles-master.svg
Profile of `loadblk` thread after changes: https://monad.io/maxopenfiles-increase.svg
Tree-SHA512: de54f77d57e9f8999eaf8d12592aab5b02f5877be8fa727a1f42cf02da2693ce25846445eb19eb138ce4e5045d1c65e14054df72faf3ff32c7655c9cfadd27a9
This commit is contained in:
parent
79639195ee
commit
89ced17da0
@ -618,7 +618,10 @@ its upstream repository.
|
||||
Current subtrees include:
|
||||
|
||||
- src/leveldb
|
||||
- Upstream at https://github.com/google/leveldb ; Maintained by Google, but open important PRs to Core to avoid delay
|
||||
- Upstream at https://github.com/google/leveldb ; Maintained by Google, but
|
||||
open important PRs to Core to avoid delay.
|
||||
- **Note**: Follow the instructions in [Upgrading LevelDB](#upgrading-leveldb) when
|
||||
merging upstream changes to the leveldb subtree.
|
||||
|
||||
- src/libsecp256k1
|
||||
- Upstream at https://github.com/bitcoin-core/secp256k1/ ; actively maintaned by Core contributors.
|
||||
@ -629,6 +632,52 @@ Current subtrees include:
|
||||
- src/univalue
|
||||
- Upstream at https://github.com/jgarzik/univalue ; report important PRs to Core to avoid delay.
|
||||
|
||||
Upgrading LevelDB
|
||||
---------------------
|
||||
|
||||
Extra care must be taken when upgrading LevelDB. This section explains issues
|
||||
you must be aware of.
|
||||
|
||||
### File Descriptor Counts
|
||||
|
||||
In most configurations we use the default LevelDB value for `max_open_files`,
|
||||
which is 1000 at the time of this writing. If LevelDB actually uses this many
|
||||
file descriptors it will cause problems with Bitcoin's `select()` loop, because
|
||||
it may cause new sockets to be created where the fd value is >= 1024. For this
|
||||
reason, on 64-bit Unix systems we rely on an internal LevelDB optimization that
|
||||
uses `mmap()` + `close()` to open table files without actually retaining
|
||||
references to the table file descriptors. If you are upgrading LevelDB, you must
|
||||
sanity check the changes to make sure that this assumption remains valid.
|
||||
|
||||
In addition to reviewing the upstream changes in `env_posix.cc`, you can use `lsof` to
|
||||
check this. For example, on Linux this command will show open `.ldb` file counts:
|
||||
|
||||
```bash
|
||||
$ lsof -p $(pidof bitcoind) |\
|
||||
awk 'BEGIN { fd=0; mem=0; } /ldb$/ { if ($4 == "mem") mem++; else fd++ } END { printf "mem = %s, fd = %s\n", mem, fd}'
|
||||
mem = 119, fd = 0
|
||||
```
|
||||
|
||||
The `mem` value shows how many files are mmap'ed, and the `fd` value shows you
|
||||
many file descriptors these files are using. You should check that `fd` is a
|
||||
small number (usually 0 on 64-bit hosts).
|
||||
|
||||
See the notes in the `SetMaxOpenFiles()` function in `dbwrapper.cc` for more
|
||||
details.
|
||||
|
||||
### Consensus Compatibility
|
||||
|
||||
It is possible for LevelDB changes to inadvertently change consensus
|
||||
compatibility between nodes. This happened in Bitcoin 0.8 (when LevelDB was
|
||||
first introduced). When upgrading LevelDB you should review the upstream changes
|
||||
to check for issues affecting consensus compatibility.
|
||||
|
||||
For example, if LevelDB had a bug that accidentally prevented a key from being
|
||||
returned in an edge case, and that bug was fixed upstream, the bug "fix" would
|
||||
be an incompatible consensus change. In this situation the correct behavior
|
||||
would be to revert the upstream fix before applying the updates to Bitcoin's
|
||||
copy of LevelDB. In general you should be wary of any upstream changes affecting
|
||||
what data is returned from LevelDB queries.
|
||||
|
||||
Git and GitHub tips
|
||||
---------------------
|
||||
|
@ -72,6 +72,31 @@ public:
|
||||
}
|
||||
};
|
||||
|
||||
static void SetMaxOpenFiles(leveldb::Options *options) {
|
||||
// On most platforms the default setting of max_open_files (which is 1000)
|
||||
// is optimal. On Windows using a large file count is OK because the handles
|
||||
// do not interfere with select() loops. On 64-bit Unix hosts this value is
|
||||
// also OK, because up to that amount LevelDB will use an mmap
|
||||
// implementation that does not use extra file descriptors (the fds are
|
||||
// closed after being mmaped).
|
||||
//
|
||||
// Increasing the value beyond the default is dangerous because LevelDB will
|
||||
// fall back to a non-mmap implementation when the file count is too large.
|
||||
// On 32-bit Unix host we should decrease the value because the handles use
|
||||
// up real fds, and we want to avoid fd exhaustion issues.
|
||||
//
|
||||
// See PR #12495 for further discussion.
|
||||
|
||||
int default_open_files = options->max_open_files;
|
||||
#ifndef WIN32
|
||||
if (sizeof(void*) < 8) {
|
||||
options->max_open_files = 64;
|
||||
}
|
||||
#endif
|
||||
LogPrint(BCLog::LEVELDB, "LevelDB using max_open_files=%d (default=%d)\n",
|
||||
options->max_open_files, default_open_files);
|
||||
}
|
||||
|
||||
static leveldb::Options GetOptions(size_t nCacheSize)
|
||||
{
|
||||
leveldb::Options options;
|
||||
@ -79,13 +104,13 @@ static leveldb::Options GetOptions(size_t nCacheSize)
|
||||
options.write_buffer_size = nCacheSize / 4; // up to two write buffers may be held in memory simultaneously
|
||||
options.filter_policy = leveldb::NewBloomFilterPolicy(10);
|
||||
options.compression = leveldb::kNoCompression;
|
||||
options.max_open_files = 64;
|
||||
options.info_log = new CBitcoinLevelDBLogger();
|
||||
if (leveldb::kMajorVersion > 1 || (leveldb::kMajorVersion == 1 && leveldb::kMinorVersion >= 16)) {
|
||||
// LevelDB versions before 1.16 consider short writes to be corruption. Only trigger error
|
||||
// on corruption in later versions.
|
||||
options.paranoid_checks = true;
|
||||
}
|
||||
SetMaxOpenFiles(&options);
|
||||
return options;
|
||||
}
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user