joellubi commented on code in PR #1456: URL: https://github.com/apache/arrow-adbc/pull/1456#discussion_r1460711856
########## go/adbc/driver/snowflake/bulk_ingestion.go: ########## @@ -0,0 +1,546 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +package snowflake + +import ( + "bytes" + "compress/flate" + "context" + "database/sql" + "database/sql/driver" + "errors" + "fmt" + "io" + "runtime" + "strings" + "sync" + + "github.com/apache/arrow-adbc/go/adbc" + "github.com/apache/arrow/go/v14/arrow" + "github.com/apache/arrow/go/v14/arrow/array" + "github.com/apache/arrow/go/v14/arrow/memory" + "github.com/apache/arrow/go/v14/parquet" + "github.com/apache/arrow/go/v14/parquet/compress" + "github.com/apache/arrow/go/v14/parquet/pqarrow" + "github.com/snowflakedb/gosnowflake" + "golang.org/x/sync/errgroup" +) + +const ( + bindStageName = "ADBC$BIND" + createTemporaryStageStmt = "CREATE OR REPLACE TEMPORARY STAGE " + bindStageName + " FILE_FORMAT = (TYPE = PARQUET USE_LOGICAL_TYPE = TRUE BINARY_AS_TEXT = FALSE)" + putQueryTmpl = "PUT 'file:///tmp/placeholder/%s' @" + bindStageName + " OVERWRITE = TRUE" + copyQuery = "COPY INTO IDENTIFIER(?) FROM @" + bindStageName + " MATCH_BY_COLUMN_NAME = CASE_INSENSITIVE" + countQuery = "SELECT COUNT(*) FROM IDENTIFIER(?)" + megabyte = 1024 * 1024 +) + +var ( + defaultTargetFileSize = 10 * megabyte + defaultWriterConcurrency = runtime.NumCPU() + defaultUploadConcurrency = 8 + defaultCopyConcurrency = 4 + defaultCompressionCodec = compress.Codecs.Snappy + defaultCompressionLevel = flate.DefaultCompression +) + +// Options for configuring bulk ingestion. +// +// Values should be updated with appropriate calls to stmt.SetOption(). +type ingestOptions struct { + // Approximate size of Parquet files written during ingestion. + // + // Actual size will be slightly larger, depending on size of footer/metadata. + // Default is 10 MB. Must be an integer > 0. + targetFileSize int + // Number of Parquet files to write in parallel. + // + // Default attempts to maximize workers based on logical cores detected, but + // may need to be adjusted if running in a constrained environment. Must be an integer > 0. + writerConcurrency int + // Number of Parquet files to upload in parallel. + // + // Greater concurrency can smooth out TCP congestion and help make use of + // available network bandwith, but will increase memory utilization. + // Default is 8. Must be an integer > 0. + uploadConcurrency int + // Maximum number of COPY operations to run concurrently. + // + // Bulk ingestion performance is optimized by executing COPY queries as files are + // still being uploaded. Snowflake COPY speed scales with warehouse size, so smaller + // warehouses may benefit from setting this value higher to ensure long-running + // COPY queries do not block newly uploaded files from being loaded. + // Default is 4. If set to 0, only a single COPY query will be executed as part of ingestion, + // once all files have finished uploading. Must be an integer >= 0. + copyConcurrency int + // Compression codec to use for Parquet files. + // + // When network speeds are high, it is generally faster to use a faster codec with + // a lower compression ratio. The opposite is true if the network is slow by CPU is + // available. + // Default is Snappy. + compressionCodec compress.Compression + // Compression level for Parquet files. + // + // The compression level is codec-specific. Some codecs do not support setting it, + // notably Snappy. + // Default is the default level for the specified compressionCodec. + compressionLevel int +} + +func DefaultIngestOptions() *ingestOptions { + return &ingestOptions{ + targetFileSize: defaultTargetFileSize, + writerConcurrency: defaultWriterConcurrency, + uploadConcurrency: defaultUploadConcurrency, + copyConcurrency: defaultCopyConcurrency, + compressionCodec: defaultCompressionCodec, + compressionLevel: defaultCompressionLevel, + } +} + +// ingestRecord performs bulk ingestion of a single Record and returns the +// number of rows affected. +// +// The Record must already be bound by calling stmt.Bind(), and will be released +// and reset upon completion. +func (st *statement) ingestRecord(ctx context.Context) (nrows int64, err error) { + defer func() { + // Record already released by writeParquet() + st.bound = nil + }() + + parquetProps, arrowProps := newWriterProps(st.alloc, st.ingestOptions) + g := errgroup.Group{} + + // writeParquet takes a channel of Records, but we only have one Record to write + recordCh := make(chan arrow.Record, 1) + recordCh <- st.bound + close(recordCh) + + // Read the Record from the channel and write it into the provided buffer + schema := st.bound.Schema() + buf := new(bytes.Buffer) + g.Go(func() error { + err = writeParquet(schema, buf, recordCh, 0, parquetProps, arrowProps) + if err != io.EOF { + return err + } + return nil + }) + + // Create a temporary stage, we can't start uploading until it has been created + _, err = st.cnxn.cn.ExecContext(ctx, createTemporaryStageStmt, nil) + if err != nil { + return + } + + // Wait for Parquet file to finish writing + err = g.Wait() + if err != nil { + return + } + + // If successful, upload the file to Snowflake + fileName := "0.parquet" // Only writing 1 file, so use same name as first file written by ingestStream() for consistency + err = uploadStream(ctx, st.cnxn.cn, buf, fileName) + if err != nil { + return + } Review Comment: Good point. I updated this. I now have the writing and stage creation kick off at the same time so that the writer buffers can fill up for whatever time we're waiting on the stage query response. After that comes back the upload starts directly from the parquet file as it's being written. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: github-unsubscr...@arrow.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
