The new InfluxDB storage engine and some query language ideas

Short talk I gave at GranfaCon

Transcript

1. The new InfluxDB storage engine and some query language ideas Paul Dix CEO at InfluxDB @pauldix paul@influxdb.com
2. preliminary intro materials…
3. Everything is indexed by time and series
4. Shards 10/11/2015 10/12/2015 Data organized into Shards of time, each is an underlying DB efficient to drop old data 10/13/2015 10/10/2015
5. InfluxDB data temperature,device=dev1,building=b1 internal=80,external=18 1443782126
6. InfluxDB data temperature,device=dev1,building=b1 internal=80,external=18 1443782126 Measurement
7. InfluxDB data temperature,device=dev1,building=b1 internal=80,external=18 1443782126 Measurement Tags
8. InfluxDB data temperature,device=dev1,building=b1 internal=80,external=18 1443782126 Measurement Tags Fields
9. InfluxDB data temperature,device=dev1,building=b1 internal=80,external=18 1443782126 Measurement Tags Fields Timestamp
10. InfluxDB data temperature,device=dev1,building=b1 internal=80,external=18 1443782126 Measurement Tags Fields Timestamp We actually store up to ns scale timestamps but I couldn’t fit on the slide
11. Each series and field to a unique ID temperature,device=dev1,building=b1#internal temperature,device=dev1,building=b1#external 1 2
12. Data per ID is tuples ordered by time temperature,device=dev1,building=b1#internal temperature,device=dev1,building=b1#external 1 2 1 (1443782126,80) 2 (1443782126,18)
13. Storage Requirements
14. High write throughput to hundreds of thousands of series
15. Awesome read performance
16. Better Compression
17. Writes can’t block reads
18. Reads can’t block writes
19. Write multiple ranges simultaneously
20. Hot backups
21. Many databases open in a single process
22. InfluxDB’s Time Structured Merge Tree (TSM Tree)
23. InfluxDB’s Time Structured Merge Tree (TSM Tree) like LSM, but different
24. Components WAL In memory cache Index Files
25. Components WAL In memory cache Index Files Similar to LSM Trees
26. Components WAL In memory cache Index Files Similar to LSM Trees Same
27. Components WAL In memory cache Index Files Similar to LSM Trees Same like MemTables
28. Components WAL In memory cache Index Files Similar to LSM Trees Same like MemTables like SSTables
29. awesome time series data WAL (an append only file)
30. awesome time series data WAL (an append only file) in memory index
31. In Memory Cache // cache and flush variables cacheLock sync.RWMutex cache map[string]Values flushCache map[string]Values temperature,device=dev1,building=b1#internal
32. In Memory Cache // cache and flush variables cacheLock sync.RWMutex cache map[string]Values flushCache map[string]Values writes can come in while WAL flushes
33. // cache and flush variables cacheLock sync.RWMutex cache map[string]Values flushCache map[string]Values dirtySort map[string]bool values can come in out of order. mark if so, sort at query time
34. Values in Memory type Value interface { Time() time.Time UnixNano() int64 Value() interface{} Size() int }
35. awesome time series data WAL (an append only file) in memory index on disk index (periodic flushes)
36. The Index Data File Min Time: 10000 Max Time: 29999 Data File Min Time: 30000 Max Time: 39999 Data File Min Time: 70000 Max Time: 99999 Contiguous blocks of time
37. The Index Data File Min Time: 10000 Max Time: 29999 Data File Min Time: 15000 Max Time: 39999 Data File Min Time: 70000 Max Time: 99999 can overlap
38. The Index cpu,host=A Min Time: 10000 Max Time: 20000 cpu,host=A Min Time: 21000 Max Time: 39999 Data File Min Time: 70000 Max Time: 99999 but a specific series must not overlap
39. The Index Data File Data File Data File a file will never overlap with more than 2 others time ascending Data File Data File
40. Data files are read only, like LSM SSTables
41. The Index Data File Min Time: 10000 Max Time: 29999 Data File Min Time: 30000 Max Time: 39999 Data File Min Time: 70000 Max Time: 99999 Data File Min Time: 10000 Max Time: 99999 they periodically get compacted (like LSM)
42. Compacting while appending new data
43. Compacting while appending new data func (w *WriteLock) LockRange(min, max int64) { // sweet code here } func (w *WriteLock) UnlockRange(min, max int64) { // sweet code here }
44. Compacting while appending new data func (w *WriteLock) LockRange(min, max int64) { // sweet code here } func (w *WriteLock) UnlockRange(min, max int64) { // sweet code here } This should block until we get it
45. Locking happens inside each Shard
46. Back to the data files… Data File Min Time: 10000 Max Time: 29999 Data File Min Time: 30000 Max Time: 39999 Data File Min Time: 70000 Max Time: 99999
47. Data File Layout
48. Data File Layout Similar to SSTables
49. Data File Layout
50. Data File Layout blocks have up to 1,000 points by default
51. Data File Layout
52. Data File Layout 4 byte position means data files can be at most 4GB
53. Data Files type dataFile struct { f *os.File size uint32 mmap []byte }
54. Memory mapping lets the OS handle caching for you
55. Compressed Data Blocks
56. Timestamps: encoding based on precision and deltas
57. Timestamps (best case): Run length encoding Deltas are all the same for a block (only requires start time, delta, and count)
58. Timestamps (good case): Simple8B Ann and Moffat in "Index compression using 64-bit words"
59. Timestamps (worst case): raw values nano-second timestamps with large deltas
60. float64: double delta Facebook’s Gorilla - google: gorilla time series facebook https://github.com/dgryski/go-tsz
61. booleans are bits!
62. int64 uses zig-zag same as from Protobufs (adding double delta and RLE)
63. string uses Snappy same compression LevelDB uses (might add dictionary compression)
64. How does it perform?
65. Compression depends greatly on the shape of your data
66. Write throughput depends on batching, CPU, and memory
67. one test: 100,000 series 100,000 points per series 10,000,000,000 total points 5,000 points per request c3.8xlarge, writes from 4 other systems ~390,000 points/sec ~3 bytes/point (random floats, could be better)
68. ~400 IOPS 30%-50% CPU There’s room for improvement!
69. Detailed writeup https://influxdb.com/docs/v0.9/concepts/storage_engine.html
70. Query Language Ideas
71. Three different kinds of functions
72. Aggregates select mean(value) from cpu where host = 'A' and time > now() - 4h group by time(5m)
73. Transformations select derivative(value) from cpu where host = 'A' and time > now() - 4h group by time(5m)
74. Selectors select min(value) from cpu where host = 'A'; and time > now() - 4h group by time(5m)
75. Then there are fills select mean(value) from cpu where host = 'A' and time > now() - 4h group by time(5m) fill(0)
76. How to differentiate between the different types?
77. How do we chain functions together? without making breaking changes to InfluxQL
78. Mix jQuery style with InfluxQL SELECT mean(value).fill(previous).derivate(1s).scale(100).as(‘mvg_avg’) FROM measurement WHERE time > now() - 4h GROUP BY time(1m)
79. D3 style SELECT mean(value) .fill(previous) .derivate(1s) .scale(100) .as(‘mvg_avg’) FROM measurement WHERE time > now() - 4h GROUP BY time(1m)
80. Moving the FROM? SELECT from('cpu').mean(value) from('memory').mean(value) WHERE time > now() - 4h GROUP BY time(1m)
81. Moving the FROM? SELECT from('cpu').mean(value) from('memory').mean(value) WHERE time > now() - 4h GROUP BY time(1m) consistent time and filtering applied to both
82. JOIN SELECT join( from('errors') .count(value), from('requests') .count(value) ).fill(0) .count(value) WHERE time > now() - 4h GROUP BY time(1m)
83. Thank you! Paul Dix @pauldix paul@influxdb.com