Fragment Metadata

Learn how to consolidate fragment metadata on your arrays.

How to run this tutorial

You can run this tutorial in two ways:

Locally on your machine.
On TileDB Cloud.

However, since TileDB Cloud has a free tier, we strongly recommend that you sign up and run everything there, as that requires no installations or deployment.

This tutorial demonstrates how to consolidate fragment metadata of arrays. Before running this tutorial, it is recommended that you read the following sections:

First, import the necessary libraries, set the array URI (i.e., its path, which in this tutorial will be on local storage), and delete any previously created arrays with the same name.

Python
R

# Import necessary libraries
import tiledb
import numpy as np
import shutil
import os.path

# Set array URI
array_uri = os.path.expanduser("~/consolidation_fragment_meta")

# Delete array if it already exists
if os.path.exists(array_uri):
    shutil.rmtree(array_uri)

# Import necessary libraries
library(tiledb)

# Set array URI
array_uri <- path.expand("~/consolidation_fragment_meta_r")

# Delete array if it already exists
if (file.exists(array_uri)) {
  unlink(array_uri, recursive = TRUE)
}

Next, create the array by specifying its schema. This example uses a sparse array, but the functionality is similar for dense arrays. Some differences in consolidation exist between sparse and dense arrays, and other sections of Academy cover those differences.

Python
R

# Create the two dimensions
d1 = tiledb.Dim(name="d1", domain=(0, 3), tile=2, dtype=np.int32)
d2 = tiledb.Dim(name="d2", domain=(0, 3), tile=2, dtype=np.int32)

# Create a domain using the two dimensions
dom = tiledb.Domain(d1, d2)

# Create an attribute
a = tiledb.Attr(name="a", dtype=np.int32)

# Create the array schema with `sparse=True`.
sch = tiledb.ArraySchema(domain=dom, sparse=True, attrs=[a])

# Create the array on disk (it will initially be empty)
tiledb.Array.create(array_uri, sch)

# Create the two dimensions
d1 <- tiledb_dim("d1", c(0L, 3L), 2L, "INT32")
d2 <- tiledb_dim("d2", c(0L, 3L), 2L, "INT32")

# Create a domain using the two dimensions
dom <- tiledb_domain(dims = c(d1, d2))

# Create an attribute
a <- tiledb_attr("a", type = "INT32")

# Create the array schema, setting `sparse = TRUE`
sch <- tiledb_array_schema(dom, a, sparse = TRUE)

# Create the array on disk (it will initially be empty)
arr <- tiledb_array_create(array_uri, sch)

Write some data to the array.

Python
R

# Prepare some data in numpy arrays
d1_data = np.array([2, 0, 3], dtype=np.int32)
d2_data = np.array([0, 1, 1], dtype=np.int32)
a_data = np.array([4, 1, 6], dtype=np.int32)

# Open the array in write mode and write the data in COO format
with tiledb.open(array_uri, "w") as A:
    A[d1_data, d2_data] = a_data

# Prepare some data in an array
d1_data <- c(2L, 0L, 3L)
d2_data <- c(0L, 1L, 1L)
a_data <- c(4L, 1L, 6L)

# Open the array for writing and write data to the array
arr <- tiledb_array(
  uri = array_uri,
  query_type = "WRITE",
  return_as = "data.frame"
)
arr[d1_data, d2_data] <- a_data

# Close the array
arr <- tiledb_array_close(arr)

Perform a second write, so that two fragments are generated.

Python
R

# Prepare some data in numpy arrays
d1_data = np.array([2, 0, 1], dtype=np.int32)
d2_data = np.array([2, 3, 3], dtype=np.int32)
a_data = np.array([5, 2, 3], dtype=np.int32)

# Open the array in write mode and write the data in COO format
with tiledb.open(array_uri, "w") as A:
    A[d1_data, d2_data] = a_data

# Prepare some data in an array
d1_data <- c(2L, 0L, 3L)
d2_data <- c(2L, 3L, 3L)
a_data <- c(5L, 2L, 3L)

# Open the array for writing and write data to the array
arr <- tiledb_array_open(
  arr,
  type = "WRITE"
)
arr[d1_data, d2_data] <- a_data

# Close the array
arr <- tiledb_array_close(arr)

The array is a folder in the path specified in array_uri. The contents are explained in other sections of the Academy, but notice that there are two fragment directories in directory __fragments, and two commit files in the __commits directory. They are all prefixed by _t1_t1 and _t2_t2, where t1 and t2 are the timestamps at which those fragments were created by the two write operations above.

/Users/stavrospapadopoulos/consolidation_fragment_meta
├── __commits
│   ├── __1716156721345_1716156721345_09dcf32eac26fa66b08d2ae4c319b0cc_21.wrt
│   └── __1716156721619_1716156721619_49f894bc1d51c8cca5d77340898797f2_21.wrt
├── __fragment_meta
├── __fragments
│   ├── __1716156721345_1716156721345_09dcf32eac26fa66b08d2ae4c319b0cc_21
│   │   ├── __fragment_metadata.tdb
│   │   ├── a0.tdb
│   │   ├── d0.tdb
│   │   └── d1.tdb
│   └── __1716156721619_1716156721619_49f894bc1d51c8cca5d77340898797f2_21
│       ├── __fragment_metadata.tdb
│       ├── a0.tdb
│       ├── d0.tdb
│       └── d1.tdb
├── __labels
├── __meta
└── __schema
    ├── __1716156719924_1716156719924_369cbcf77cbf43734ffb77cb0fcb2c5a
    └── __enumerations

10 directories, 11 files

Read the array data. Observe that the data from both writes are present in the result, as expected.

Python
R

# Read array
with tiledb.open(array_uri, "r") as A:
    print(A[:])

OrderedDict({'a': array([1, 2, 3, 4, 6, 5], dtype=int32), 'd1': array([0, 0, 1, 2, 3, 2], dtype=int32), 'd2': array([1, 3, 3, 0, 1, 2], dtype=int32)})

# Open the array in read mode
arr <- tiledb_array_open(arr, type = "READ")

# Show the entire array
cat("Entire array:\n")
print(arr[])

arr <- tiledb_array_close(arr)

Entire array:
  d1 d2 a
1  0  1 1
2  2  0 4
3  3  1 6
4  0  3 2
5  2  2 5
6  3  3 3

Now consolidate the array, passing fragment_meta as the value of configuration parameter sm.consolidation.mode.

Python
R

# Consolidate
config = tiledb.Config({"sm.consolidation.mode": "fragment_meta"})
tiledb.consolidate(array_uri, config=config)

# Consolidate
cfg <- tiledb_config(
  config = c(
    "sm.consolidation.mode" = "fragment_meta"
  )
)

array_consolidate(array_uri, cfg = cfg)

Inspecting the file hierarchy of the array again, observe that TileDB created a new file inside the __fragment_meta directory, prefixed by _t1_t2 and with suffix .meta. This contains information about the metadata of both fragments present in the array.

/Users/stavrospapadopoulos/consolidation_fragment_meta
├── __commits
│   ├── __1716156721345_1716156721345_09dcf32eac26fa66b08d2ae4c319b0cc_21.wrt
│   └── __1716156721619_1716156721619_49f894bc1d51c8cca5d77340898797f2_21.wrt
├── __fragment_meta
│   └── __1716156721345_1716156721619_788367824bd1620e8f4ad739b8075314_21.meta
├── __fragments
│   ├── __1716156721345_1716156721345_09dcf32eac26fa66b08d2ae4c319b0cc_21
│   │   ├── __fragment_metadata.tdb
│   │   ├── a0.tdb
│   │   ├── d0.tdb
│   │   └── d1.tdb
│   └── __1716156721619_1716156721619_49f894bc1d51c8cca5d77340898797f2_21
│       ├── __fragment_metadata.tdb
│       ├── a0.tdb
│       ├── d0.tdb
│       └── d1.tdb
├── __labels
├── __meta
└── __schema
    ├── __1716156719924_1716156719924_369cbcf77cbf43734ffb77cb0fcb2c5a
    └── __enumerations

10 directories, 12 files

Read the array again and observe that nothing changed. The same results are present there. Consolidation (in combination with vacuuming), helps significantly boost performance in the presence of multiple write operations. For more details, visit the Performance: Tuning Writes section.

Python
R

# Read array
with tiledb.open(array_uri, "r") as A:
    print(A[:])

OrderedDict({'a': array([1, 2, 3, 4, 6, 5], dtype=int32), 'd1': array([0, 0, 1, 2, 3, 2], dtype=int32), 'd2': array([1, 3, 3, 0, 1, 2], dtype=int32)})

# Open the array in read mode
arr <- tiledb_array_open(arr, type = "READ")

# Show the entire array
cat("Entire array:\n")
print(arr[])

arr <- tiledb_array_close(arr)

Entire array:
  d1 d2 a
1  0  1 1
2  2  0 4
3  3  1 6
4  0  3 2
5  2  2 5
6  3  3 3

Clean up in the end by deleting the array.

Python
R

# Delete the array
if os.path.exists(array_uri):
    shutil.rmtree(array_uri)

# Delete the array
if (file.exists(array_uri)) {
  unlink(array_uri, recursive = TRUE)
}

Note

For more advanced configuration for consolidation, visit the Key Concepts: Consolidation section.