bulk-post 0.2.0__tar.gz → 0.3.0__tar.gz

{bulk_post-0.2.0/src/bulk_post.egg-info → bulk_post-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: bulk-post
-Version: 0.2.0
+Version: 0.3.0
 Summary: A near-stdlib Python CLI that fires templated HTTP requests driven by CSV data, one request (or a multi-step workflow) per row.
 Author-email: Valerii Kirichenko <true.monte.kristo@gmail.com>
 License-Expression: MIT
@@ -31,7 +31,7 @@ Dynamic: license-file
 [![Python](https://img.shields.io/badge/python-3.12%2B-blue.svg)](https://www.python.org/downloads/)
 [![Security: report a vulnerability](https://img.shields.io/badge/security-report%20a%20vulnerability-red.svg)](https://github.com/true-monte-kristo/bulk-post/security/advisories/new)
 
-A near-stdlib Python CLI that fires templated HTTP requests driven by CSV data. You define the request — URL, method, body, headers — with `{{placeholder}}` slots, and each CSV row supplies the values that fill them: one request per row, or a multi-step request workflow per row in `--workflow` mode. Supports bearer or basic auth (default: no auth) with automatic 401 re-prompt, a live terminal UI with pause/resume, parallel execution, and a retry file for failed rows. Third-party dependencies: PyYAML and jsonpath-ng — both always installed, but lazily imported (PyYAML only on the `--workflow` code path, jsonpath-ng only on the workflow-variables code path).
+A near-stdlib Python CLI that fires templated HTTP requests driven by CSV data. You define the request — URL, method, body, headers — with `{{placeholder}}` slots, and each CSV row supplies the values that fill them: one request per row, or a multi-step request workflow per row in `--workflow` mode. Supports bearer or basic auth (default: no auth) with automatic 401 re-prompt, a live terminal UI with pause/resume, parallel execution, and a redrive file for failed rows. Third-party dependencies: PyYAML and jsonpath-ng — both always installed, but lazily imported (PyYAML only on the `--workflow` code path, jsonpath-ng only on the workflow-variables code path).
 
 ## Requirements
 
@@ -137,20 +137,26 @@ bulk-post -u "https://api.example.com/items/{{id}}" -c items.csv -o 47
 | `--delay` | `-d` | `0` | Milliseconds to wait between requests |
 | `--offset` | `-o` | `0` | Skip the first N data rows (useful for resuming after a failure) |
 | `--timeout` | `-T` | `30` | Per-request timeout in seconds |
-| `--retry-file` | `-r` | `<stem>_failed.csv` | Where to write rows that failed; auto-named from the CSV path if omitted |
+| `--redrive-file` | `-r` | `<stem>_failed.csv` | Where to write rows that failed; auto-named from the CSV path if omitted |
 | `--verbose` | `-v` | false | Print URL, request/response headers (Authorization masked), body, status, and timing for every row |
 | `--header` | `-H` | — | Add a custom request header in `Name: value` format; repeatable. Values support `{{col}}` placeholders |
 | `--parallel` | `-p` | false | Process rows concurrently using multiple threads; `--delay` is ignored in this mode |
 | `--concurrency-level` | `-n` | CPU count | Number of worker threads; only used with `--parallel` |
 | `--debug` | `-D` | false | Print worker thread name on each row log line and show a live debug bar with queue depth, active thread count, and ok/fail counters; only meaningful with `--parallel` |
 | `--workflow` | `-w` | — | Path to a workflow YAML file; mutually exclusive with `--url` |
+| `--retry-on` | `-R` | — (off) | Comma-separated HTTP status codes to retry on (e.g. `503,429`); enables retries |
+| `--retry-backoff` | `-B` | `fixed` | `fixed` or `exponential`; requires `--retry-on` |
+| `--max-retries` | `-M` | `5` | Retries after the initial attempt (5 ⇒ up to 6 requests); requires `--retry-on` |
+| `--retry-delay` | `-y` | `200` | Milliseconds; fixed delay, or initial delay for exponential; requires `--retry-on` |
+| `--multiplier` | `-x` | `1.5` | Exponential backoff multiplier; requires `--retry-backoff exponential` |
+| `--max-retry-delay` | `-Y` | `30000` | Millisecond hard cap on any single retry wait; requires `--retry-on` |
 | `--version` | `-V` | — | Print version and exit |
 
 ## CSV format
 
 The CSV must have a header row. Column names are used as placeholder names in `--url`, `--body`, and `--header` values. Every `{{placeholder}}` in the URL, body, or header values must match a column name; the script exits with an error if any are missing.
 
-The input delimiter is detected automatically (comma, semicolon, tab, or pipe), falling back to comma when it can't be determined. The failed-rows retry CSV is written with the same delimiter as the input. There is no delimiter flag.
+The input delimiter is detected automatically (comma, semicolon, tab, or pipe), falling back to comma when it can't be determined. The failed-rows redrive CSV is written with the same delimiter as the input. There is no delimiter flag.
 
 ```csv
 id,reason
@@ -176,6 +182,33 @@ Pass `--auth-type basic` (or `-a basic`). Credentials (`user:pass`) are resolved
 
 The default when `--auth-type` is omitted (or pass `--auth-type none` / `-a none` explicitly). No `Authorization` header is sent.
 
+## Retries
+
+Retries are disabled by default. Pass `--retry-on` with a comma-separated list of HTTP status codes to enable them:
+
+```bash
+bulk-post -u "https://api.example.com/{{id}}" -c rows.csv -R 503,429 -B exponential
+```
+
+Only the exact listed status codes trigger a retry — network errors and timeouts do not. If the server returns a `Retry-After` or `X-Retry-After` header, the wait is extended to at least that value before the next attempt. All waits are capped by `--max-retry-delay`. Each retry prints a short `[RETRY]` notice; `--verbose` adds the elapsed time and body snippet of the failed attempt.
+
+**Flags:**
+
+| Flag | Short | Default | Description |
+|------|-------|---------|-------------|
+| `--retry-on` | `-R` | — | Comma-separated status codes (e.g. `503,429`); required to enable retries |
+| `--retry-backoff` | `-B` | `fixed` | `fixed` or `exponential` |
+| `--max-retries` | `-M` | `5` | Retries after the first attempt (5 ⇒ up to 6 total requests) |
+| `--retry-delay` | `-y` | `200` | ms; fixed wait between retries, or starting delay for exponential |
+| `--multiplier` | `-x` | `1.5` | Growth factor for exponential backoff |
+| `--max-retry-delay` | `-Y` | `30000` | ms hard cap on any single retry wait |
+
+**Notes:**
+
+- `401` cannot be listed in `--retry-on` when `--auth-type` is `bearer` or `basic` — the 401 auth-refresh flow owns that status code.
+- Retry sleeps are interruptible: `/pause` freezes the countdown, `/exit` abandons the wait and routes the row to the redrive file.
+- In workflow mode, retry flags are not available; use `retry_policy:` in the workflow YAML instead (see [Workflow mode](#workflow-mode)).
+
 ## Terminal UI
 
 When running in an interactive terminal, a live bottom bar shows:
@@ -193,11 +226,11 @@ Available commands:
 
 In non-TTY mode (piped input, CI, test environments) the bottom bar is skipped and no interactive commands are available.
 
-## Retry file
+## Redrive file
 
-Rows that fail (network error, non-2xx response, or substitution error) are written to the retry file. By default this is `<csv-stem>_failed.csv` next to the input file. Re-run with `-c <stem>_failed.csv` to retry only those rows.
+Rows that fail (network error, non-2xx response, or substitution error) are written to the redrive file. By default this is `<csv-stem>_failed.csv` next to the input file. Re-run with `-c <stem>_failed.csv` to redrive only those rows.
 
-If no rows fail, the retry file is deleted automatically.
+If no rows fail, the redrive file is deleted automatically.
 
 ## Workflow mode
 
@@ -209,6 +242,7 @@ Each CSV row fires all steps in document order. Steps within a row are always se
 
 ```yaml
 workflow:
+  persist_context: false           # optional; default false — see "Redrive and resume" below
   description: Optional human-readable description  # skipped at runtime
 
   groupA:                          # logical grouping for shared auth
@@ -249,12 +283,60 @@ Key rules:
 
 - **Execution order** — steps fire in the order they appear in the document (top to bottom across all groups).
 - **Group auth** — all steps in a group inherit the group's auth unless they declare their own.
-- **`on_error`** — `stop` (default) halts remaining steps for that row and writes it to the retry file; `continue` logs the failure, writes the row, and proceeds to the next step.
+- **`on_error`** — `stop` (default) halts remaining steps for that row and writes it to the redrive file; `continue` logs the failure, writes the row, and proceeds to the next step.
 - **Placeholders** — `{{col}}` in `url`, `body`, and header values is replaced with the matching CSV column value, same as in single-URL mode.
 
-### Retry and resume
+### Workflow retries
+
+Each group and each endpoint may carry an optional `retry_policy:` block. An endpoint-level `retry_policy` **replaces** the group's entirely — there is no merging. `retry_policy: none` on an endpoint disables any inherited group policy. `on_error` evaluates only the final post-retries result.
+
+```yaml
+workflow:
+  groupA:
+    auth:
+      type: bearer
+    retry_policy:           # group-level policy, inherited by all endpoints in the group
+      retry_on: 503,429     # required: comma-separated string or YAML list
+      backoff: exponential  # optional: fixed (default) or exponential
+      multiplier: 1.5       # optional: exponential only; defaults to 1.5
+      delay: 200            # optional: fixed delay / initial delay in ms; defaults to 200
+      max_retries: 5        # optional: retries after the initial attempt; defaults to 5
+      max_delay: 30000      # optional: hard cap on any single retry wait in ms; defaults to 30000
+    endpoints:
+      - call-api:
+          url: https://api.example.com/{{id}}
+          method: POST
+
+  groupB:
+    endpoints:
+      - call-other:
+          url: https://other.example.com/{{id}}
+          method: DELETE
+          retry_policy:     # endpoint-level policy replaces any group policy entirely
+            retry_on: 503,429
+            backoff: fixed
+            delay: 200
+            max_retries: 5
+
+  groupC:
+    retry_policy:           # group has a policy …
+      retry_on: 503
+    endpoints:
+      - no-retry-step:
+          retry_policy: none  # … but this endpoint opts out
+          url: https://public.example.com/{{id}}
+          method: GET
+```
+
+### Redrive and resume
+
+Mid-workflow resume is opt-in via a `persist_context: true` key at the top of the `workflow:` mapping (default `false`).
+
+- **`persist_context: false` (default):** failed rows are written to the redrive file with only the original input columns. Re-running always starts each row from step 1. If the input CSV contains `_bulk_post_step` or `_bulk_post_var/…` context columns (e.g. from a previous run with `persist_context: true`), they are ignored, with a startup warning printed to stderr.
+
+- **`persist_context: true`:** when a step fails, the row is written to the redrive file with an extra column `_bulk_post_step` set to the path of the first failed step (e.g. `groupA/step-name`) plus any persisted variable columns. Re-running with that redrive CSV skips all steps before the failed one, resuming mid-workflow automatically.
 
-When a step fails, the row is written to the retry file with an extra column `_bulk_post_step` set to the path of the first failed step (e.g. `groupA/step-name`). Re-running with that retry CSV skips all steps before the failed one, resuming mid-workflow automatically.
+> **Security note:** `persist_context: true` writes response-derived data (potentially sensitive) to disk in plaintext. Do not share or commit redrive CSVs produced with this option enabled.
 
 ### Workflow variables
 
@@ -291,15 +373,15 @@ workflow:
 - Names must start with `$` (e.g. `$id`) and are referenced as `{{$id}}` in URL, headers, and body.
 - `source` is written as `.workflow.<group>.<endpoint>` (or just `<group>/<endpoint>`). It must refer to an endpoint that runs before the current step — forward and self references are rejected at startup.
 - `jsonPath` uses full JSONPath syntax (powered by [`jsonpath-ng`](https://pypi.org/project/jsonpath-ng/)). Only the first match is used. A match that is an object or array (non-scalar) fails the step.
-- `nullable` defaults to `true`. When `false`, a null value or no-match fails the step (row written to retry file); when `true`, it resolves to an empty string.
+- `nullable` defaults to `true`. When `false`, a null value or no-match fails the step (row written to redrive file); when `true`, it resolves to an empty string.
 - Variable values are scoped to a single CSV row and are never shared across rows.
 - All variable declarations are validated at startup — undefined references, bad names, unreachable sources, and invalid JSONPath expressions all cause an immediate exit with a clear error.
 
-**Resume/retry with variables:**
+**Resume/redrive with variables:**
 
-When a row fails, any resolved variable values are persisted into reserved retry-CSV columns named `_bulk_post_var/<source_path>/<name>`. Re-running the retry CSV skips completed steps and reads these persisted values for variables whose source step was skipped.
+Requires `persist_context: true` at the top of the `workflow:` mapping. When enabled, on row failure any resolved variable values are persisted into reserved redrive-CSV columns named `_bulk_post_var/<source_path>/<name>`. Re-running the redrive CSV skips completed steps and reads these persisted values for variables whose source step was skipped. When `persist_context` is `false` (the default), variable values are not persisted and re-runs always start from step 1.
 
-> **Security note:** retry CSVs may contain response-derived data (potentially sensitive) in plaintext. Do not share or commit retry CSVs that were produced from workflows using variables.
+> **Security note:** redrive CSVs may contain response-derived data (potentially sensitive) in plaintext. Do not share or commit redrive CSVs that were produced from workflows using variables.
 
 ### Example
 

{bulk_post-0.2.0 → bulk_post-0.3.0}/README.md

@@ -5,7 +5,7 @@
 [![Python](https://img.shields.io/badge/python-3.12%2B-blue.svg)](https://www.python.org/downloads/)
 [![Security: report a vulnerability](https://img.shields.io/badge/security-report%20a%20vulnerability-red.svg)](https://github.com/true-monte-kristo/bulk-post/security/advisories/new)
 
-A near-stdlib Python CLI that fires templated HTTP requests driven by CSV data. You define the request — URL, method, body, headers — with `{{placeholder}}` slots, and each CSV row supplies the values that fill them: one request per row, or a multi-step request workflow per row in `--workflow` mode. Supports bearer or basic auth (default: no auth) with automatic 401 re-prompt, a live terminal UI with pause/resume, parallel execution, and a retry file for failed rows. Third-party dependencies: PyYAML and jsonpath-ng — both always installed, but lazily imported (PyYAML only on the `--workflow` code path, jsonpath-ng only on the workflow-variables code path).
+A near-stdlib Python CLI that fires templated HTTP requests driven by CSV data. You define the request — URL, method, body, headers — with `{{placeholder}}` slots, and each CSV row supplies the values that fill them: one request per row, or a multi-step request workflow per row in `--workflow` mode. Supports bearer or basic auth (default: no auth) with automatic 401 re-prompt, a live terminal UI with pause/resume, parallel execution, and a redrive file for failed rows. Third-party dependencies: PyYAML and jsonpath-ng — both always installed, but lazily imported (PyYAML only on the `--workflow` code path, jsonpath-ng only on the workflow-variables code path).
 
 ## Requirements
 
@@ -111,20 +111,26 @@ bulk-post -u "https://api.example.com/items/{{id}}" -c items.csv -o 47
 | `--delay` | `-d` | `0` | Milliseconds to wait between requests |
 | `--offset` | `-o` | `0` | Skip the first N data rows (useful for resuming after a failure) |
 | `--timeout` | `-T` | `30` | Per-request timeout in seconds |
-| `--retry-file` | `-r` | `<stem>_failed.csv` | Where to write rows that failed; auto-named from the CSV path if omitted |
+| `--redrive-file` | `-r` | `<stem>_failed.csv` | Where to write rows that failed; auto-named from the CSV path if omitted |
 | `--verbose` | `-v` | false | Print URL, request/response headers (Authorization masked), body, status, and timing for every row |
 | `--header` | `-H` | — | Add a custom request header in `Name: value` format; repeatable. Values support `{{col}}` placeholders |
 | `--parallel` | `-p` | false | Process rows concurrently using multiple threads; `--delay` is ignored in this mode |
 | `--concurrency-level` | `-n` | CPU count | Number of worker threads; only used with `--parallel` |
 | `--debug` | `-D` | false | Print worker thread name on each row log line and show a live debug bar with queue depth, active thread count, and ok/fail counters; only meaningful with `--parallel` |
 | `--workflow` | `-w` | — | Path to a workflow YAML file; mutually exclusive with `--url` |
+| `--retry-on` | `-R` | — (off) | Comma-separated HTTP status codes to retry on (e.g. `503,429`); enables retries |
+| `--retry-backoff` | `-B` | `fixed` | `fixed` or `exponential`; requires `--retry-on` |
+| `--max-retries` | `-M` | `5` | Retries after the initial attempt (5 ⇒ up to 6 requests); requires `--retry-on` |
+| `--retry-delay` | `-y` | `200` | Milliseconds; fixed delay, or initial delay for exponential; requires `--retry-on` |
+| `--multiplier` | `-x` | `1.5` | Exponential backoff multiplier; requires `--retry-backoff exponential` |
+| `--max-retry-delay` | `-Y` | `30000` | Millisecond hard cap on any single retry wait; requires `--retry-on` |
 | `--version` | `-V` | — | Print version and exit |
 
 ## CSV format
 
 The CSV must have a header row. Column names are used as placeholder names in `--url`, `--body`, and `--header` values. Every `{{placeholder}}` in the URL, body, or header values must match a column name; the script exits with an error if any are missing.
 
-The input delimiter is detected automatically (comma, semicolon, tab, or pipe), falling back to comma when it can't be determined. The failed-rows retry CSV is written with the same delimiter as the input. There is no delimiter flag.
+The input delimiter is detected automatically (comma, semicolon, tab, or pipe), falling back to comma when it can't be determined. The failed-rows redrive CSV is written with the same delimiter as the input. There is no delimiter flag.
 
 ```csv
 id,reason
@@ -150,6 +156,33 @@ Pass `--auth-type basic` (or `-a basic`). Credentials (`user:pass`) are resolved
 
 The default when `--auth-type` is omitted (or pass `--auth-type none` / `-a none` explicitly). No `Authorization` header is sent.
 
+## Retries
+
+Retries are disabled by default. Pass `--retry-on` with a comma-separated list of HTTP status codes to enable them:
+
+```bash
+bulk-post -u "https://api.example.com/{{id}}" -c rows.csv -R 503,429 -B exponential
+```
+
+Only the exact listed status codes trigger a retry — network errors and timeouts do not. If the server returns a `Retry-After` or `X-Retry-After` header, the wait is extended to at least that value before the next attempt. All waits are capped by `--max-retry-delay`. Each retry prints a short `[RETRY]` notice; `--verbose` adds the elapsed time and body snippet of the failed attempt.
+
+**Flags:**
+
+| Flag | Short | Default | Description |
+|------|-------|---------|-------------|
+| `--retry-on` | `-R` | — | Comma-separated status codes (e.g. `503,429`); required to enable retries |
+| `--retry-backoff` | `-B` | `fixed` | `fixed` or `exponential` |
+| `--max-retries` | `-M` | `5` | Retries after the first attempt (5 ⇒ up to 6 total requests) |
+| `--retry-delay` | `-y` | `200` | ms; fixed wait between retries, or starting delay for exponential |
+| `--multiplier` | `-x` | `1.5` | Growth factor for exponential backoff |
+| `--max-retry-delay` | `-Y` | `30000` | ms hard cap on any single retry wait |
+
+**Notes:**
+
+- `401` cannot be listed in `--retry-on` when `--auth-type` is `bearer` or `basic` — the 401 auth-refresh flow owns that status code.
+- Retry sleeps are interruptible: `/pause` freezes the countdown, `/exit` abandons the wait and routes the row to the redrive file.
+- In workflow mode, retry flags are not available; use `retry_policy:` in the workflow YAML instead (see [Workflow mode](#workflow-mode)).
+
 ## Terminal UI
 
 When running in an interactive terminal, a live bottom bar shows:
@@ -167,11 +200,11 @@ Available commands:
 
 In non-TTY mode (piped input, CI, test environments) the bottom bar is skipped and no interactive commands are available.
 
-## Retry file
+## Redrive file
 
-Rows that fail (network error, non-2xx response, or substitution error) are written to the retry file. By default this is `<csv-stem>_failed.csv` next to the input file. Re-run with `-c <stem>_failed.csv` to retry only those rows.
+Rows that fail (network error, non-2xx response, or substitution error) are written to the redrive file. By default this is `<csv-stem>_failed.csv` next to the input file. Re-run with `-c <stem>_failed.csv` to redrive only those rows.
 
-If no rows fail, the retry file is deleted automatically.
+If no rows fail, the redrive file is deleted automatically.
 
 ## Workflow mode
 
@@ -183,6 +216,7 @@ Each CSV row fires all steps in document order. Steps within a row are always se
 
 ```yaml
 workflow:
+  persist_context: false           # optional; default false — see "Redrive and resume" below
   description: Optional human-readable description  # skipped at runtime
 
   groupA:                          # logical grouping for shared auth
@@ -223,12 +257,60 @@ Key rules:
 
 - **Execution order** — steps fire in the order they appear in the document (top to bottom across all groups).
 - **Group auth** — all steps in a group inherit the group's auth unless they declare their own.
-- **`on_error`** — `stop` (default) halts remaining steps for that row and writes it to the retry file; `continue` logs the failure, writes the row, and proceeds to the next step.
+- **`on_error`** — `stop` (default) halts remaining steps for that row and writes it to the redrive file; `continue` logs the failure, writes the row, and proceeds to the next step.
 - **Placeholders** — `{{col}}` in `url`, `body`, and header values is replaced with the matching CSV column value, same as in single-URL mode.
 
-### Retry and resume
+### Workflow retries
+
+Each group and each endpoint may carry an optional `retry_policy:` block. An endpoint-level `retry_policy` **replaces** the group's entirely — there is no merging. `retry_policy: none` on an endpoint disables any inherited group policy. `on_error` evaluates only the final post-retries result.
+
+```yaml
+workflow:
+  groupA:
+    auth:
+      type: bearer
+    retry_policy:           # group-level policy, inherited by all endpoints in the group
+      retry_on: 503,429     # required: comma-separated string or YAML list
+      backoff: exponential  # optional: fixed (default) or exponential
+      multiplier: 1.5       # optional: exponential only; defaults to 1.5
+      delay: 200            # optional: fixed delay / initial delay in ms; defaults to 200
+      max_retries: 5        # optional: retries after the initial attempt; defaults to 5
+      max_delay: 30000      # optional: hard cap on any single retry wait in ms; defaults to 30000
+    endpoints:
+      - call-api:
+          url: https://api.example.com/{{id}}
+          method: POST
+
+  groupB:
+    endpoints:
+      - call-other:
+          url: https://other.example.com/{{id}}
+          method: DELETE
+          retry_policy:     # endpoint-level policy replaces any group policy entirely
+            retry_on: 503,429
+            backoff: fixed
+            delay: 200
+            max_retries: 5
+
+  groupC:
+    retry_policy:           # group has a policy …
+      retry_on: 503
+    endpoints:
+      - no-retry-step:
+          retry_policy: none  # … but this endpoint opts out
+          url: https://public.example.com/{{id}}
+          method: GET
+```
+
+### Redrive and resume
+
+Mid-workflow resume is opt-in via a `persist_context: true` key at the top of the `workflow:` mapping (default `false`).
+
+- **`persist_context: false` (default):** failed rows are written to the redrive file with only the original input columns. Re-running always starts each row from step 1. If the input CSV contains `_bulk_post_step` or `_bulk_post_var/…` context columns (e.g. from a previous run with `persist_context: true`), they are ignored, with a startup warning printed to stderr.
+
+- **`persist_context: true`:** when a step fails, the row is written to the redrive file with an extra column `_bulk_post_step` set to the path of the first failed step (e.g. `groupA/step-name`) plus any persisted variable columns. Re-running with that redrive CSV skips all steps before the failed one, resuming mid-workflow automatically.
 
-When a step fails, the row is written to the retry file with an extra column `_bulk_post_step` set to the path of the first failed step (e.g. `groupA/step-name`). Re-running with that retry CSV skips all steps before the failed one, resuming mid-workflow automatically.
+> **Security note:** `persist_context: true` writes response-derived data (potentially sensitive) to disk in plaintext. Do not share or commit redrive CSVs produced with this option enabled.
 
 ### Workflow variables
 
@@ -265,15 +347,15 @@ workflow:
 - Names must start with `$` (e.g. `$id`) and are referenced as `{{$id}}` in URL, headers, and body.
 - `source` is written as `.workflow.<group>.<endpoint>` (or just `<group>/<endpoint>`). It must refer to an endpoint that runs before the current step — forward and self references are rejected at startup.
 - `jsonPath` uses full JSONPath syntax (powered by [`jsonpath-ng`](https://pypi.org/project/jsonpath-ng/)). Only the first match is used. A match that is an object or array (non-scalar) fails the step.
-- `nullable` defaults to `true`. When `false`, a null value or no-match fails the step (row written to retry file); when `true`, it resolves to an empty string.
+- `nullable` defaults to `true`. When `false`, a null value or no-match fails the step (row written to redrive file); when `true`, it resolves to an empty string.
 - Variable values are scoped to a single CSV row and are never shared across rows.
 - All variable declarations are validated at startup — undefined references, bad names, unreachable sources, and invalid JSONPath expressions all cause an immediate exit with a clear error.
 
-**Resume/retry with variables:**
+**Resume/redrive with variables:**
 
-When a row fails, any resolved variable values are persisted into reserved retry-CSV columns named `_bulk_post_var/<source_path>/<name>`. Re-running the retry CSV skips completed steps and reads these persisted values for variables whose source step was skipped.
+Requires `persist_context: true` at the top of the `workflow:` mapping. When enabled, on row failure any resolved variable values are persisted into reserved redrive-CSV columns named `_bulk_post_var/<source_path>/<name>`. Re-running the redrive CSV skips completed steps and reads these persisted values for variables whose source step was skipped. When `persist_context` is `false` (the default), variable values are not persisted and re-runs always start from step 1.
 
-> **Security note:** retry CSVs may contain response-derived data (potentially sensitive) in plaintext. Do not share or commit retry CSVs that were produced from workflows using variables.
+> **Security note:** redrive CSVs may contain response-derived data (potentially sensitive) in plaintext. Do not share or commit redrive CSVs that were produced from workflows using variables.
 
 ### Example
 

{bulk_post-0.2.0 → bulk_post-0.3.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "bulk-post"
-version = "0.2.0"
+version = "0.3.0"
 description = "A near-stdlib Python CLI that fires templated HTTP requests driven by CSV data, one request (or a multi-step workflow) per row."
 readme = "README.md"
 requires-python = ">=3.12"

{bulk_post-0.2.0 → bulk_post-0.3.0}/src/bulk_post/__init__.py

@@ -58,7 +58,7 @@ from .csvio import (
     _open_log_file as _open_log_file,
 )
 from .csvio import (
-    _open_retry_writer as _open_retry_writer,
+    _open_redrive_writer as _open_redrive_writer,
 )
 from .csvio import (
     _skip_rows as _skip_rows,
@@ -74,6 +74,18 @@ from .csvio import (
 )
 from .http import _mask_headers as _mask_headers
 from .http import http_request as http_request
+from .retry import (
+    RetryPolicy as RetryPolicy,
+)
+from .retry import (
+    compute_delay as compute_delay,
+)
+from .retry import (
+    parse_retry_policy as parse_retry_policy,
+)
+from .retry import (
+    request_with_retry as request_with_retry,
+)
 from .runner import (
     _fire as _fire,
 )
@@ -212,6 +224,9 @@ from .workflow import _WORKFLOW_STEP_COL as _WORKFLOW_STEP_COL
 from .workflow import (
     WorkflowStep as WorkflowStep,
 )
+from .workflow import (
+    WorkflowSteps as WorkflowSteps,
+)
 from .workflow import (
     _fire_workflow_step as _fire_workflow_step,
 )